[dev.typealias] all: merge go1.8.3 into dev.typealias

352996a381 (tag: go1.8.3) [release-branch.go1.8] go1.8.3
bb5055d6f1 [release-branch.go1.8] doc: document go1.8.3
439c0c8be8 [release-branch.go1.8] cmd/compile: don't move spills to loop exits where the spill is dead
e396667ba3 [release-branch.go1.8] cmd/compile: zero ambiguously live variables at VARKILLs
daf6706f37 [release-branch.go1.8] runtime: use pselect6 for usleep on linux/386
958c64bbab [release-branch.go1.8] runtime: use pselect6 for usleep on linux/amd64 and linux/arm
195e20a976 [release-branch.go1.8] cmd/compile: ignore types when considering tuple select for CSE
f55bc1c4eb [release-branch.go1.8] net/http: update bundled http2 for gracefulShutdownCh lock contention slowdown
51f508bb4a [release-branch.go1.8] cmd/compile: fix s390x unsigned comparison constant merging rules
243dee1737 [release-branch.go1.8] cmd/go: if we get a C compiler dwarf2 warning, try without -g
a43c0d2dc8 [release-branch.go1.8] runtime: don't corrupt arena bounds on low mmap
1054085dcf [release-branch.go1.8] cmd/compile: fix store chain in schedule pass
18a13d373a [release-branch.go1.8] runtime: doubly fix "double wakeup" panic
6efa2f22ac [release-branch.go1.8] database/sql: ensure releaseConn is defined before a possible close
fb9770f09b [release-branch.go1.8] runtime: print debug info on "base out of range"
b6a8fc8d8c [release-branch.go1.8] doc: remove mentions of yacc tool
59870f9e19 (tag: go1.8.2) [release-branch.go1.8] go1.8.2
c9688ddb6b [release-branch.go1.8] doc: document go1.8.2 and go1.7.6
38d35f49e7 [release-branch.go1.8] crypto/elliptic: fix carry bug in x86-64 P-256 implementation.

Change-Id: I2aa0eab7a990d24e25809fb13ce6cb031104f474

README.md

@@ -5,39 +5,37 @@ reliable, and efficient software.
 
 ![Gopher image](doc/gopher/fiveyears.jpg)
 
-For documentation about how to install and use Go,
-visit https://golang.org/ or load doc/install-source.html
-in your web browser.
-
 Our canonical Git repository is located at https://go.googlesource.com/go.
 There is a mirror of the repository at https://github.com/golang/go.
 
+Unless otherwise noted, the Go source files are distributed under the
+BSD-style license found in the LICENSE file.
+
+### Download and Install
+
+#### Binary Distributions
+
+Official binary distributions are available at https://golang.org/dl/.
+
+After downloading a binary release, visit https://golang.org/doc/install
+or load doc/install.html in your web browser for installation
+instructions.
+
+#### Install From Source
+
+If a binary distribution is not available for your combination of
+operating system and architecture, visit
+https://golang.org/doc/install/source or load doc/install-source.html
+in your web browser for source installation instructions.
+
+### Contributing
+
 Go is the work of hundreds of contributors. We appreciate your help!
 
 To contribute, please read the contribution guidelines:
 	https://golang.org/doc/contribute.html
 
-##### Note that we do not accept pull requests and that we use the issue tracker for bug reports and proposals only. Please ask questions on https://forum.golangbridge.org or https://groups.google.com/forum/#!forum/golang-nuts.
-
-Unless otherwise noted, the Go source files are distributed
-under the BSD-style license found in the LICENSE file.
-
---
-
-## Binary Distribution Notes
-
-If you have just untarred a binary Go distribution, you need to set
-the environment variable $GOROOT to the full path of the go
-directory (the one containing this file).  You can omit the
-variable if you unpack it into /usr/local/go, or if you rebuild
-from sources by running all.bash (see doc/install-source.html).
-You should also add the Go binary directory $GOROOT/bin
-to your shell's path.
-
-For example, if you extracted the tar file into $HOME/go, you might
-put the following in your .profile:
-
-	export GOROOT=$HOME/go
-	export PATH=$PATH:$GOROOT/bin
-
-See https://golang.org/doc/install or doc/install.html for more details.
+Note that the Go project does not use GitHub pull requests, and that
+we use the issue tracker for bug reports and proposals only. See
+https://golang.org/wiki/Questions for a list of places to ask
+questions about the Go language.

VERSION

@@ -0,0 +1 @@
+go1.8.3.typealias

doc/cmd.html

@@ -22,8 +22,6 @@ using the go <code>tool</code> subcommand, such as <code>go tool vet</code>.
 This style of invocation allows, for instance, checking a single source file
 rather than an entire package: <code>go tool vet myprogram.go</code> as
 compared to <code>go vet mypackage</code>.
-Some of the commands, such as <code>yacc</code>, are accessible only through
-the go <code>tool</code> subcommand.
 </p>
 
 <p>
@@ -95,12 +93,6 @@ gofmt</a> command with more general options.</td>
 calls whose arguments do not align with the format string.</td>
 </tr>
 
-<tr>
-<td><a href="/cmd/yacc/">yacc</a></td>
-<td>&nbsp;&nbsp;&nbsp;&nbsp;</td>
-<td>Yacc is a version of yacc that generates parsers implemented in Go.</td>
-</tr>
-
 </table>
 
 <p>

doc/conduct.html

@@ -148,29 +148,26 @@ These actions are explicitly forbidden in Go spaces:
 
 <p>
 The Go spaces are not free speech venues; they are for discussion about Go.
-These spaces have moderators.
-The goal of the moderators is to facilitate civil discussion about Go.
+Each of these spaces have their own moderators.
 </p>
 
 <p>
 When using the official Go spaces you should act in the spirit of the “Gopher
 values”.
-If you conduct yourself in a way that is explicitly forbidden by the CoC,
-you will be warned and asked to stop.
-If you do not stop, you will be removed from our community spaces temporarily.
-Repeated, willful breaches of the CoC will result in a permanent ban.
+If a reported conflict cannot be resolved amicably, the CoC Working Group
+may make a recommendation to the relevant forum moderators.
 </p>
 
 <p>
-Moderators are held to a higher standard than other community members.
-If a moderator creates an inappropriate situation, they should expect less
-leeway than others, and should expect to be removed from their position if they
-cannot adhere to the CoC.
+CoC Working Group members and forum moderators are held to a higher standard than other community members.
+If a working group member or moderator creates an inappropriate situation, they
+should expect less leeway than others, and should expect to be removed from
+their position if they cannot adhere to the CoC.
 </p>
 
 <p>
-Complaints about moderator actions must be handled using the reporting process
-below.
+Complaints about working group member or moderator actions must be handled 
+using the reporting process below.
 </p>
 
 <h2 id="reporting">Reporting issues</h2>
@@ -185,8 +182,6 @@ satisfaction of all parties. They are:
 <ul>
 	<li>Aditya Mukerjee &lt;dev@chimeracoder.net&gt;
 	<li>Andrew Gerrand &lt;adg@golang.org&gt;
-	<li>Dave Cheney &lt;dave@cheney.net&gt;
-	<li>Jason Buberel &lt;jbuberel@google.com&gt;
 	<li>Peggy Li &lt;peggyli.224@gmail.com&gt;
 	<li>Sarah Adams &lt;sadams.codes@gmail.com&gt;
 	<li>Steve Francia &lt;steve.francia@gmail.com&gt;
@@ -201,13 +196,10 @@ particular individual or group.
 </p>
 
 <ul>
-<li>Mail <a href="mailto:conduct@golang.org">conduct@golang.org</a> or
-    <a href="https://golang.org/s/conduct-report">submit an anonymous report</a>.
+<li>Mail <a href="mailto:conduct@golang.org">conduct@golang.org</a>.
     <ul>
     <li>Your message will reach the Working Group.
     <li>Reports are confidential within the Working Group.
-    <li>Should you choose to remain anonymous then the Working Group cannot
-        notify you of the outcome of your report.
     <li>You may contact a member of the group directly if you do not feel
         comfortable contacting the group as a whole. That member will then raise
         the issue with the Working Group as a whole, preserving the privacy of the
@@ -229,11 +221,8 @@ particular individual or group.
 <li>The Working Group will reach a decision as to how to act. These may include:
     <ul>
     <li>Nothing.
-    <li>A request for a private or public apology.
-    <li>A private or public warning.
-    <li>An imposed vacation (for instance, asking someone to abstain for a week
-        from a mailing list or IRC).
-    <li>A permanent or temporary ban from some or all Go spaces.
+    <li>Passing the report along to the offender.
+    <li>A recommendation of action to the relevant forum moderators.
     </ul>
 <li>The Working Group will reach out to the original reporter to let them know
     the decision.
@@ -246,7 +235,6 @@ particular individual or group.
 conflicts in the most harmonious way possible.</b>
 We hope that in most cases issues may be resolved through polite discussion and
 mutual agreement.
-Bannings and other forceful measures are to be employed only as a last resort.
 </p>
 
 <p>

doc/contrib.html

@@ -34,6 +34,7 @@ We encourage all Go users to subscribe to
 <p>A <a href="/doc/devel/release.html">summary</a> of the changes between Go releases. Notes for the major releases:</p>
 
 <ul>
+	<li><a href="/doc/go1.8">Go 1.8</a> <small>(February 2017)</small></li>
 	<li><a href="/doc/go1.7">Go 1.7</a> <small>(August 2016)</small></li>
 	<li><a href="/doc/go1.6">Go 1.6</a> <small>(February 2016)</small></li>
 	<li><a href="/doc/go1.5">Go 1.5</a> <small>(August 2015)</small></li>

doc/contribute.html

@@ -1,14 +1,12 @@
 <!--{
-	"Title": "Contribution Guidelines"
+	"Title": "Contribution Guide"
 }-->
 
-<h2 id="Introduction">Introduction</h2>
-
 <p>
-This document explains how to contribute changes to the Go project.
-It assumes you have followed the
-<a href="/doc/install/source">installation instructions</a> and
-have <a href="code.html">written and tested your code</a>.
+The Go project welcomes all contributors. The process of contributing
+to the Go project may be different than many projects you are used to.
+This document is intended as a guide to help you through the contribution
+process. This guide assumes you have a basic understanding of Git and Go.
 </p>
 
 <p>
@@ -20,103 +18,54 @@ see <a href="gccgo_contribute.html">Contributing to gccgo</a>.)
 Sensitive security-related issues should be reported to <a href="mailto:security@golang.org">security@golang.org</a>.
 </p>
 
-<h2 id="Design">Discuss your design</h2>
+<h1 id="contributor">Becoming a contributor</h1>
 
 <p>
-The project welcomes submissions but please let everyone know what
-you're working on if you want to change or add to the Go repositories.
+Before you can contribute to the Go project you need to setup a few prerequisites.
+The Go project uses <a href="https://www.gerritcodereview.com/">Gerrit</a>, an open
+source online tool, to perform all code reviews.
+Gerrit uses your email address as a unique identifier.
+The Go project contributing flow is currently configured to work only with Google Accounts.
+You must go through the following process <em>prior to contributing</em>.
+You only need to do this once per Google Account.
+</p>
+
+<h2 id="auth">Configure Git to use Gerrit</h2>
+<p>
+You'll need a web browser and a command line terminal.
+You should already have Git installed.
 </p>
 
 <p>
-Before undertaking to write something new for the Go project,
-please <a href="https://golang.org/issue/new">file an issue</a>
-(or claim an <a href="https://golang.org/issues">existing issue</a>).
-Significant changes must go through the
-<a href="https://golang.org/s/proposal-process">change proposal process</a>
-before they can be accepted.
-</p>
-
-<p>
-This process gives everyone a chance to validate the design,
-helps prevent duplication of effort,
-and ensures that the idea fits inside the goals for the language and tools.
-It also checks that the design is sound before code is written;
-the code review tool is not the place for high-level discussions.
-</p>
-
-<p>
-When planning work, please note that the Go project follows a
-<a href="https://golang.org/wiki/Go-Release-Cycle">six-month
-development cycle</a>. The latter half of each cycle is a three-month
-feature freeze during which only bug fixes and doc updates are accepted.
-New work cannot be submitted during a feature freeze.
-</p>
-
-<h2 id="Testing">Testing redux</h2>
-
-<p>
-You've <a href="code.html">written and tested your code</a>, but
-before sending code out for review, run all the tests for the whole
-tree to make sure the changes don't break other packages or programs:
-</p>
-
-<pre>
-$ cd go/src
-$ ./all.bash
-</pre>
-
-<p>
-(To build under Windows use <code>all.bat</code>.)
-</p>
-
-<p>
-After running for a while, the command should print
-"<code>ALL</code> <code>TESTS</code> <code>PASSED</code>".
-</p>
-
-<h2 id="Code_review">Code review</h2>
-
-<p>
-Changes to Go must be reviewed before they are accepted,
-no matter who makes the change.
-A custom git command called <code>git-codereview</code>,
-discussed below, helps manage the code review process through a Google-hosted
-<a href="https://go-review.googlesource.com/">instance</a> of the code review
-system called <a href="https://www.gerritcodereview.com/">Gerrit</a>.
-</p>
-
-<h3 id="auth">Set up authentication for code review</h3>
-
-<p>
-Gerrit uses Google Accounts for authentication. If you don't have
-a Google Account, you can create an account which
+Gerrit uses Google Accounts for authentication.
+If you don't have a Google Account, you can create an account which
 <a href="https://www.google.com/accounts/NewAccount">includes
 a new Gmail email account</a> or create an account associated
 <a href="https://accounts.google.com/SignUpWithoutGmail">with your existing
 email address</a>.
 </p>
 
-<p>
-The email address associated with the Google Account you use will be recorded in
-the <a href="https://go.googlesource.com/go/+log/">change log</a>
-and in the <a href="/CONTRIBUTORS">contributors file</a>.
-</p>
+<h3>Step 1: Sign in to googlesource and generate a password</h3>
 
 <p>
-To set up your account in Gerrit, visit
-<a href="https://go.googlesource.com">go.googlesource.com</a>
+Visit <a href="https://go.googlesource.com">go.googlesource.com</a>
 and click on "Generate Password" in the page's top right menu bar.
-</p>
-
-<p>
 You will be redirected to accounts.google.com to sign in.
 </p>
 
+<h3>Step 2: Run the provided script</h3>
 <p>
-Once signed in, you are returned back to go.googlesource.com to "Configure Git".
-Follow the instructions on the page.
-(If you are on a Windows computer, you should instead follow the instructions
-in the yellow box to run the command.)
+After signing in, you are taken to a page on go.googlesource.com with the title "Configure Git".
+This page contains a personalized script which when run locally will configure git
+to have your unique authentication key.
+This key is paired with one generated server side similar to how ssh keys work.
+</p>
+
+<p>
+Copy and run this script locally in your command line terminal.
+(On a Windows computer using cmd you should instead follow the instructions
+in the yellow box to run the command. If you are using git-bash use the same
+script as *nix.)
 </p>
 
 <p>
@@ -124,23 +73,25 @@ Your secret authentication token is now in a <code>.gitcookie</code> file
 and Git is configured to use this file.
 </p>
 
-<h3 id="gerrit">Register with Gerrit</h3>
+<h3 id="gerrit">Step 3: Register with Gerrit</h3>
 
 <p>
-Now that you have your authentication token,
-you need to register your account with Gerrit.
-To do this, visit
-<a href="https://go-review.googlesource.com/login/">
-go-review.googlesource.com/login/</a>. You will immediately be redirected
-to Google Accounts. Sign in using the same Google Account you used above.
-That is all that is required.
+Now that you have your authentication token, you need to register your
+account with Gerrit.
+To do this, visit <a href="https://go-review.googlesource.com/login/">
+go-review.googlesource.com/login/</a>.
+Sign in using the same Google Account you used above.
 </p>
 
-<h3 id="cla">Contributor License Agreement</h3>
+<h2 id="cla">Contributor License Agreement</h2>
+
+<h3 id="which_cla">Which CLA</h3>
+<p>
+Before sending your first change to the Go project
+you must have completed one of the following two CLAs.
+Which CLA you should sign depends on who owns the copyright to your work.
+</p>
 
-<p>Gerrit serves as the gatekeeper and uses your e-mail address as the key.
-To send your first change to the Go project from a given address,
-you must have completed one of the contributor license agreements:
 <ul>
 <li>
 If you are the copyright holder, you will need to agree to the
@@ -151,37 +102,49 @@ contributor license agreement</a>, which can be completed online.
 If your organization is the copyright holder, the organization
 will need to agree to the
 <a href="https://developers.google.com/open-source/cla/corporate">corporate
-contributor license agreement</a>.
-(If the copyright holder for your code has already completed the
-agreement in connection with another Google open source project,
-it does not need to be completed again.)
+contributor license agreement</a>.<br>
 </li>
 </ul>
 
 <p>
-You can use the links above to create and sign the contributor license agreement
-or you can show your current agreements and create new ones through the Gerrit
-interface.  <a href="https://go-review.googlesource.com/login/">Log into Gerrit</a>,
+<i>If the copyright holder for your contribution has already completed the
+agreement in connection with another Google open source project,
+it does not need to be completed again.</i>
+</p>
+
+<h3 id="signing_cla">Completing the CLA</h3>
+
+<p>
+You can see your currently signed agreements and sign new ones through the Gerrit
+interface.
+To do this, <a href="https://go-review.googlesource.com/login/">Log into Gerrit</a>,
 click your name in the upper-right, choose "Settings", then select "Agreements"
-from the topics on the left. If you do not have a signed agreement listed here,
+from the topics on the left.
+If you do not have a signed agreement listed here,
 you can create one by clicking "New Contributor Agreement" and following the steps.
 </p>
 
 <p>
-This rigmarole only needs to be done for your first submission for each email address.
+If the copyright holder for the code you are submitting changes &mdash; for example,
+if you start contributing code on behalf of a new company &mdash; please send email
+to golang-dev and let us know, so that we can make sure an appropriate agreement is
+completed and update the <code>AUTHORS</code> file.
 </p>
 
+<span id="Code_review"></span>
+<h1 id="prepare_dev_env">Preparing a Development Environment for Contributing</h1>
+
+<h2 id="git-codereview">Setting up Git for submission to Gerrit</h2>
 <p>
-If the copyright holder for the code you are submitting changes—for example,
-if you start contributing code on behalf of a new company—please send email
-to let us know, so that we can make sure an appropriate agreement is completed
-and update the <code>AUTHORS</code> file.
+Changes to Go must be reviewed before they are accepted, no matter who makes the change.
+A custom git command called <code>git-codereview</code>, discussed below,
+helps manage the code review process through a Google-hosted
+<a href="https://go-review.googlesource.com/">instance</a> Gerrit.
 </p>
 
-<h3 id="git-codereview">Install the git-codereview command</h3>
-
+<h3 id="git-codereview_install">Install the git-codereview command</h3>
 <p>
-Now install the <code>git-codereview</code> command by running,
+Install the <code>git-codereview</code> command by running,
 </p>
 
 <pre>
@@ -202,18 +165,28 @@ prints help text, not an error.
 </p>
 
 <p>
-<b>Note to Git aficionados:</b>
-The <code>git-codereview</code> command is not required to
-upload and manage Gerrit code reviews. For those who prefer plain Git, the text
-below gives the Git equivalent of each git-codereview command.
+On Windows, when using git-bash you must make sure that
+<code>git-codereview.exe</code> is in your git exec-path.
+Run <code>git --exec-path</code> to discover the right location then create a
+symbolic link or simply copy the executible from $GOPATH/bin to this directory.
 </p>
 
-<p>If you do use plain
-Git, note that you still need the commit hooks that the git-codereview command
-configures; those hooks add a Gerrit <code>Change-Id</code> line to the commit
-message and check that all Go source files have been formatted with gofmt. Even
-if you intend to use plain Git for daily work, install the hooks in a new Git
-checkout by running <code>git-codereview</code> <code>hooks</code>.
+<p>
+<b>Note to Git aficionados:</b>
+The <code>git-codereview</code> command is not required to
+upload and manage Gerrit code reviews.
+For those who prefer plain Git, the text below gives the Git equivalent of
+each git-codereview command.
+</p>
+
+<p>
+If you do use plain Git, note that you still need the commit hooks that the
+git-codereview command configures; those hooks add a Gerrit
+<code>Change-Id</code> line to the commit message and check that all Go source
+files have been formatted with gofmt.
+Even if you intend to use plain Git for
+daily work, install the hooks in a new Git checkout by running
+<code>git-codereview</code> <code>hooks</code>.
 </p>
 
 <p>
@@ -264,7 +237,8 @@ To install them, copy this text into your Git configuration file
 	sync = codereview sync
 </pre>
 
-<h3 id="help">Understanding the git-codereview command</h3>
+<span id="help"></span>
+<h3 id="understanding_git-codereview">Understanding the git-codereview command</h3>
 
 <p>After installing the <code>git-codereview</code> command, you can run</p>
 
@@ -277,11 +251,70 @@ to learn more about its commands.
 You can also read the <a href="https://godoc.org/golang.org/x/review/git-codereview">command documentation</a>.
 </p>
 
-<h3 id="master">Switch to the master branch</h3>
+
+<h1 id="making_a_contribution">Making a Contribution</h1>
+
+<h2 id="Design">Discuss your design</h2>
+
+<p>
+The project welcomes submissions but please let everyone know what
+you're working on if you want to change or add to the Go repositories.
+</p>
+
+<p>
+Before undertaking to write something new for the Go project,
+please <a href="https://golang.org/issue/new">file an issue</a>
+(or claim an <a href="https://golang.org/issues">existing issue</a>).
+Significant changes must go through the
+<a href="https://golang.org/s/proposal-process">change proposal process</a>
+before they can be accepted.
+</p>
+
+<p>
+This process gives everyone a chance to validate the design,
+helps prevent duplication of effort,
+and ensures that the idea fits inside the goals for the language and tools.
+It also checks that the design is sound before code is written;
+the code review tool is not the place for high-level discussions.
+</p>
+
+<p>
+When planning work, please note that the Go project follows a <a
+href="https://golang.org/wiki/Go-Release-Cycle">six-month development cycle</a>.
+The latter half of each cycle is a three-month feature freeze during
+which only bug fixes and doc updates are accepted. New contributions can be
+sent during a feature freeze but will not be accepted until the freeze thaws.
+</p>
+
+<h2 id="making_a_change">Making a change</h2>
+
+<h3 id="checkout_go">Getting Go Source</h3>
+<p>
+First you need to have a local copy of the source checked out from the correct
+repository.
+As Go builds Go you will also likely need to have a working version
+of Go installed (some documentation changes may not need this).
+This should be a recent version of Go and can be obtained via any package or
+binary distribution or you can build it from source.
+</p>
+
+<p>
+You should checkout the Go source repo anywhere you want as long as it's
+outside of your $GOPATH.
+Go to a directory where you want the source to appear and run the following
+command in a terminal.
+</p>
+
+<pre><code>
+$ git clone https://go.googlesource.com/go
+$ cd go
+</code></pre>
+
+<h3 id="master">Contributing to the main Go tree</h3>
 
 <p>
 Most Go installations use a release branch, but new changes should
-only be made based on the master branch.
+only be made based on the master branch. <br>
 (They may be applied later to a release branch as part of the release process,
 but most contributors won't do this themselves.)
 Before making a change, make sure you start on the master branch:
@@ -297,10 +330,61 @@ $ git sync
 <code>git</code> <code>pull</code> <code>-r</code>.)
 </p>
 
-<h3 id="change">Make a change</h3>
+<h3 id="subrepos">Contributing to subrepositories (golang.org/x/...)</h3>
+
+<p>
+If you are contributing a change to a subrepository, obtain the
+Go package using <code>go get</code>. For example, to contribute
+to <code>golang.org/x/oauth2</code>, check out the code by running:
+</p>
+
+<pre>
+$ go get -d golang.org/x/oauth2/...
+</pre>
+
+<p>
+Then, change your directory to the package's source directory
+(<code>$GOPATH/src/golang.org/x/oauth2</code>).
+</p>
+
+<h3 id="change">Make your changes</h3>
+
+<p>
+The entire checked-out tree is editable.
+Make your changes as you see fit ensuring that you create appropriate
+tests along with your changes. Test your changes as you go.
+</p>
+
+<h3 id="copyright">Copyright</h3>
+
+<p>
+Files in the Go repository don't list author names, both to avoid clutter
+and to avoid having to keep the lists up to date.
+Instead, your name will appear in the
+<a href="https://golang.org/change">change log</a> and in the <a
+href="/CONTRIBUTORS"><code>CONTRIBUTORS</code></a> file and perhaps the <a
+href="/AUTHORS"><code>AUTHORS</code></a> file.
+These files are automatically generated from the commit logs perodically.
+The <a href="/AUTHORS"><code>AUTHORS</code></a> file defines who &ldquo;The Go
+Authors&rdquo;&mdash;the copyright holders&mdash;are.
+</p>
+
+<p>New files that you contribute should use the standard copyright header:</p>
+
+<pre>
+// Copyright 2017 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+</pre>
+
+<p>
+Files in the repository are copyright the year they are added.
+Do not update the copyright year on files that you change.
+</p>
+
+<h3 id="commit_changes">Commit your changes</h3>
 
 <p>
-The entire checked-out tree is writable.
 Once you have edited files, you must tell Git that they have been modified.
 You must also tell Git about any files that are added, removed, or renamed files.
 These operations are done with the usual Git commands,
@@ -311,16 +395,19 @@ and
 </p>
 
 <p>
-If you wish to checkpoint your work, or are ready to send the code out for review, run</p>
+Once you have the changes queued up, you will want to commit them.
+In the Go contribution workflow this is done with a `git change` command,
+which creates a local branch and commits the changes directly to that local branch.
+</p>
 
 <pre>
 $ git change <i>&lt;branch&gt;</i>
 </pre>
 
 <p>
-from any directory in your Go repository to commit the changes so far.
 The name <i>&lt;branch&gt;</i> is an arbitrary one you choose to identify the
-local branch containing your changes.
+local branch containing your changes and will not be used elsewhere.
+This is an offline operation and nothing will be sent to the server yet.
 </p>
 
 <p>
@@ -331,9 +418,11 @@ then <code>git</code> <code>commit</code>.)
 </p>
 
 <p>
-Git will open a change description file in your editor.
+As the `git commit` is the final step, Git will open an editor to ask for a
+commit message.
 (It uses the editor named by the <code>$EDITOR</code> environment variable,
 <code>vi</code> by default.)
+
 The file will look like:
 </p>
 
@@ -352,7 +441,7 @@ At the beginning of this file is a blank line; replace it
 with a thorough description of your change.
 The first line of the change description is conventionally a one-line
 summary of the change, prefixed by the primary affected package,
-and is used as the subject for code review mail.
+and is used as the subject for code review email.
 It should complete the sentence "This change modifies Go to _____."
 The rest of the description elaborates and should provide context for the
 change and explain what it does.
@@ -395,7 +484,7 @@ the command and move that file to a different branch.
 <p>
 The special notation "Fixes #159" associates the change with issue 159 in the
 <a href="https://golang.org/issue/159">Go issue tracker</a>.
-When this change is eventually submitted, the issue
+When this change is eventually applied, the issue
 tracker will automatically mark the issue as fixed.
 (There are several such conventions, described in detail in the
 <a href="https://help.github.com/articles/closing-issues-via-commit-messages/">GitHub Issue Tracker documentation</a>.)
@@ -406,6 +495,13 @@ Once you have finished writing the commit message,
 save the file and exit the editor.
 </p>
 
+<p>
+You must have the $EDITOR environment variable set properly and working properly (exiting cleanly)
+for this operation to succeed.
+If you run into any issues at this step, it's likely your editor isn't exiting cleanly.
+Try setting a different editor in your $EDITOR environment variable.
+</p>
+
 <p>
 If you wish to do more editing, re-stage your changes using
 <code>git</code> <code>add</code>, and then run
@@ -416,8 +512,8 @@ $ git change
 </pre>
 
 <p>
-to update the change description and incorporate the staged changes.  The
-change description contains a <code>Change-Id</code> line near the bottom,
+to update the change description and incorporate the staged changes.
+The change description contains a <code>Change-Id</code> line near the bottom,
 added by a Git commit hook during the initial
 <code>git</code> <code>change</code>.
 That line is used by Gerrit to match successive uploads of the same change.
@@ -429,35 +525,44 @@ Do not edit or delete it.
 runs <code>git</code> <code>commit</code> <code>--amend</code>.)
 </p>
 
-<h3 id="mail">Mail the change for review</h3>
+<h3 id="Testing">Testing</h3>
 
 <p>
-Once the change is ready, mail it out for review:
+You've <a href="code.html">written and tested your code</a>, but
+before sending code out for review, run all the tests for the whole
+tree to make sure the changes don't break other packages or programs:
+</p>
+
+<pre>
+$ cd go/src
+$ ./all.bash
+</pre>
+
+<p>
+(To build under Windows use <code>all.bat</code>.)
+</p>
+
+<p>
+After running for a while, the command should print
+</p>
+
+<pre>
+"ALL TESTS PASSED".
+</pre>
+
+<h3 id="mail">Send the change for review</h3>
+
+<p>
+Once the change is ready, send it for review.
+This is similar to a <code>git push</code> in a GitHub style workflow.
+This is done via the mail alias setup earlier which despite its name, doesn't
+directly mail anything, it simply sends the change to Gerrit via git push.
 </p>
 
 <pre>
 $ git mail
 </pre>
 
-<p>
-You can specify a reviewer or CC interested parties
-using the <code>-r</code> or <code>-cc</code> options.
-Both accept a comma-separated list of email addresses:
-</p>
-
-<pre>
-$ git mail -r joe@golang.org -cc mabel@example.com,math-nuts@swtch.com
-</pre>
-
-<p>
-Unless explicitly told otherwise, such as in the discussion leading
-up to sending in the change list, it's better not to specify a reviewer.
-All changes are automatically CC'ed to the
-<a href="https://groups.google.com/group/golang-codereviews">golang-codereviews@googlegroups.com</a>
-mailing list. If this is your first ever change, there may be a moderation
-delay before it appears on the mailing list, to prevent spam.
-</p>
-
 <p>
 (In Git terms, <code>git</code> <code>mail</code> pushes the local committed
 changes to Gerrit using <code>git</code> <code>push</code> <code>origin</code>
@@ -479,7 +584,76 @@ remote: New Changes:
 remote:   https://go-review.googlesource.com/99999 math: improved Sin, Cos and Tan precision for very large arguments
 </pre>
 
-<h3 id="review">Reviewing code</h3>
+<h3>Troubleshooting</h3>
+
+<p>
+The most common way that the <code>git mail</code> command fails is because the
+email address used has not gone through the setup above.
+<br>
+If you see something like...
+</p>
+
+<pre>
+remote: Processing changes: refs: 1, done
+remote:
+remote: ERROR:  In commit ab13517fa29487dcf8b0d48916c51639426c5ee9
+remote: ERROR:  author email address XXXXXXXXXXXXXXXXXXX
+remote: ERROR:  does not match your user account.
+</pre>
+
+<p>
+You need to either add the email address listed to the CLA or set this repo to use
+another email address already approved.
+</p>
+
+<p>
+First let's change the email address for this repo so this doesn't happen again.
+You can change your email address for this repo with the following command:
+</p>
+
+<pre>
+$ git config user.email email@address.com
+</pre>
+
+<p>
+Then change the previous commit to use this alternative email address.
+You can do that with:
+</p>
+
+<pre>
+$ git commit --amend --author="Author Name &lt;email@address.com&gt;"
+</pre>
+
+<p>
+Finally try to resend with:
+</p>
+
+<pre>
+$ git mail
+</pre>
+
+<h3 id="cc">Specifying a reviewer / CCing others</h3>
+
+<p>
+Unless explicitly told otherwise, such as in the discussion leading
+up to sending in the change list, it's better not to specify a reviewer.
+All changes are automatically CC'ed to the
+<a href="https://groups.google.com/group/golang-codereviews">golang-codereviews@googlegroups.com</a>
+mailing list. If this is your first ever change, there may be a moderation
+delay before it appears on the mailing list, to prevent spam.
+</p>
+
+<p>
+You can specify a reviewer or CC interested parties
+using the <code>-r</code> or <code>-cc</code> options.
+Both accept a comma-separated list of email addresses:
+</p>
+
+<pre>
+$ git mail -r joe@golang.org -cc mabel@example.com,math-nuts@swtch.com
+</pre>
+
+<h2 id="review">Going through the review process</h2>
 
 <p>
 Running <code>git</code> <code>mail</code> will send an email to you and the
@@ -491,7 +665,15 @@ You must reply through the web interface.
 (Unlike with the old Rietveld review system, replying by mail has no effect.)
 </p>
 
-<h3 id="revise">Revise and upload</h3>
+<h3 id="revise">Revise and resend</h3>
+
+<p>
+The Go contribution workflow is optimized for iterative revisions based on
+feedback.
+It is rare that an initial contribution will be ready to be applied as is.
+As you revise your contribution and resend Gerrit will retain a history of
+all the changes and comments made in the single URL.
+</p>
 
 <p>
 You must respond to review comments through the web interface.
@@ -534,6 +716,8 @@ $ git sync
 <code>git</code> <code>pull</code> <code>-r</code>.)
 </p>
 
+<h3 id="resolving_conflicts">Resolving Conflicts</h3>
+
 <p>
 If files you were editing have changed, Git does its best to merge the
 remote changes into your local changes.
@@ -609,8 +793,8 @@ might turn up:
 <p>
 Git doesn't show it, but suppose the original text that both edits
 started with was 1e8; you changed it to 1e10 and the other change to 1e9,
-so the correct answer might now be 1e10.  First, edit the section
-to remove the markers and leave the correct code:
+so the correct answer might now be 1e10.
+First, edit the section to remove the markers and leave the correct code:
 </p>
 
 <pre>
@@ -639,10 +823,13 @@ restore the change commit.
 <h3 id="download">Reviewing code by others</h3>
 
 <p>
-You can import a change proposed by someone else into your local Git repository.
+As part of the review process reviewers can propose changes directly (in the
+GitHub workflow this would be someone else attaching commits to a pull request).
+
+You can import these changes proposed by someone else into your local Git repository.
 On the Gerrit review page, click the "Download ▼" link in the upper right
-corner, copy the "Checkout" command and run it from your local Git repo.
-It should look something like this:
+corner, copy the "Checkout" command and run it from your local Git repo. It
+should look something like this:
 </p>
 
 <pre>
@@ -653,11 +840,11 @@ $ git fetch https://go.googlesource.com/review refs/changes/21/1221/1 &amp;&amp;
 To revert, change back to the branch you were working in.
 </p>
 
-<h3 id="submit">Submit the change after the review</h3>
+<h2 id="submit">Apply the change to the master branch</h2>
 
 <p>
 After the code has been <code>LGTM</code>'ed, an approver may
-submit it to the master branch using the Gerrit UI.
+apply it to the master branch using the Gerrit UI.
 There is a "Submit" button on the web page for the change
 that appears once the change is approved (marked +2).
 </p>
@@ -669,41 +856,13 @@ and the code review will be updated with a link to the change
 in the repository.
 Since the method used to integrate the changes is "Cherry Pick",
 the commit hashes in the repository will be changed by
-the submit operation.
+the "Submit" operation.
 </p>
 
-<h3 id="more">More information</h3>
+<h2 id="more">More information</h2>
 
 <p>
-In addition to the information here, the Go community maintains a <a href="https://golang.org/wiki/CodeReview">CodeReview</a> wiki page.
+In addition to the information here, the Go community maintains a <a
+href="https://golang.org/wiki/CodeReview">CodeReview</a> wiki page.
 Feel free to contribute to this page as you learn the review process.
 </p>
-
-<h2 id="copyright">Copyright</h2>
-
-<p>Files in the Go repository don't list author names,
-both to avoid clutter and to avoid having to keep the lists up to date.
-Instead, your name will appear in the
-<a href="https://golang.org/change">change log</a>
-and in the <a href="/CONTRIBUTORS"><code>CONTRIBUTORS</code></a> file
-and perhaps the <a href="/AUTHORS"><code>AUTHORS</code></a> file.
-</p>
-
-<p>The <a href="/CONTRIBUTORS"><code>CONTRIBUTORS</code></a> file
-defines who the Go contributors&mdash;the people&mdash;are;
-the <a href="/AUTHORS"><code>AUTHORS</code></a> file defines
-who &ldquo;The Go Authors&rdquo;&mdash;the copyright holders&mdash;are.
-These files will be periodically updated based on the commit logs.
-
-<p>Code that you contribute should use the standard copyright header:</p>
-
-<pre>
-// Copyright 2016 The Go Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-</pre>
-
-<p>
-Files in the repository are copyright the year they are added. It is not
-necessary to update the copyright year on files that you change.
-</p>

doc/debugging_with_gdb.html

@@ -4,7 +4,8 @@
 }-->
 
 <p><i>
-This applies to the <code>gc</code> toolchain. Gccgo has native gdb support.
+This applies to the standard toolchain (the <code>gc</code> Go
+compiler and tools). Gccgo has native gdb support.
 Besides this overview you might want to consult the
 <a href="http://sourceware.org/gdb/current/onlinedocs/gdb/">GDB manual</a>.
 </i></p>
@@ -49,6 +50,14 @@ when debugging, pass the flags <code>-gcflags "-N -l"</code> to the
 debugged.
 </p>
 
+<p>
+If you want to use gdb to inspect a core dump, you can trigger a dump
+on a program crash, on systems that permit it, by setting
+<code>GOTRACEBACK=crash</code> in the environment (see the
+<a href="/pkg/runtime/#hdr-Environment_Variables"> runtime package
+documentation</a> for more info).
+</p>
+
 <h3 id="Common_Operations">Common Operations</h3>
 
 <ul>
@@ -130,7 +139,7 @@ the DWARF code.
 
 <p>
 If you're interested in what the debugging information looks like, run
-'<code>objdump -W 6.out</code>' and browse through the <code>.debug_*</code>
+'<code>objdump -W a.out</code>' and browse through the <code>.debug_*</code>
 sections.
 </p>
 
@@ -377,7 +386,9 @@ $3 = struct hchan&lt;*testing.T&gt;
 </pre>
 
 <p>
-That <code>struct hchan&lt;*testing.T&gt;</code> is the runtime-internal representation of a channel.  It is currently empty, or gdb would have pretty-printed it's contents.
+That <code>struct hchan&lt;*testing.T&gt;</code> is the
+runtime-internal representation of a channel. It is currently empty,
+or gdb would have pretty-printed its contents.
 </p>
 
 <p>

doc/devel/release.html

@@ -30,6 +30,39 @@ to fix critical security problems in both Go 1.4 and Go 1.5 as they arise.
 See the <a href="/security">security policy</a> for more details.
 </p>
 
+<h2 id="go1.8">go1.8 (released 2017/02/16)</h2>
+
+<p>
+Go 1.8 is a major release of Go.
+Read the <a href="/doc/go1.8">Go 1.8 Release Notes</a> for more information.
+</p>
+
+<h3 id="go1.8.minor">Minor revisions</h3>
+
+<p>
+go1.8.1 (released 2017/04/07) includes fixes to the compiler, linker, runtime,
+documentation, <code>go</code> command and the <code>crypto/tls</code>,
+<code>encoding/xml</code>, <code>image/png</code>, <code>net</code>,
+<code>net/http</code>, <code>reflect</code>, <code>text/template</code>,
+and <code>time</code> packages.
+See the <a href="https://github.com/golang/go/issues?q=milestone%3AGo1.8.1">Go
+1.8.1 milestone</a> on our issue tracker for details.
+</p>
+
+<p>
+go1.8.2 (released 2017/05/23) includes a security fix to the
+<code>crypto/elliptic</code> package.
+See the <a href="https://github.com/golang/go/issues?q=milestone%3AGo1.8.2">Go
+1.8.2 milestone</a> on our issue tracker for details.
+</p>
+
+<p>
+go1.8.3 (released 2017/05/24) includes fixes to the compiler, runtime,
+documentation, and the <code>database/sql</code> package.
+See the <a href="https://github.com/golang/go/issues?q=milestone%3AGo1.8.3">Go
+1.8.3 milestone</a> on our issue tracker for details.
+</p>
+
 <h2 id="go1.7">go1.7 (released 2016/08/15)</h2>
 
 <p>
@@ -69,6 +102,20 @@ See the <a href="https://github.com/golang/go/issues?q=milestone%3AGo1.7.4">Go
 1.7.4 milestone</a> on our issue tracker for details.
 </p>
 
+<p>
+go1.7.5 (released 2017/01/26) includes fixes to the compiler, runtime,
+and the <code>crypto/x509</code> and <code>time</code> packages.
+See the <a href="https://github.com/golang/go/issues?q=milestone%3AGo1.7.5">Go
+1.7.5 milestone</a> on our issue tracker for details.
+</p>
+
+<p>
+go1.7.6 (released 2017/05/23) includes the same security fix as Go 1.8.2 and
+was released at the same time.
+See the <a href="https://github.com/golang/go/issues?q=milestone%3AGo1.8.2">Go
+1.8.2 milestone</a> on our issue tracker for details.
+</p>
+
 <h2 id="go1.6">go1.6 (released 2016/02/17)</h2>
 
 <p>

doc/gccgo_install.html

@@ -52,6 +52,19 @@ user libraries.  The Go 1.4 runtime is not fully merged, but that
 should not be visible to Go programs.
 </p>
 
+<p>
+The GCC 6 releases include a complete implementation of the Go 1.6.1
+user libraries.  The Go 1.6 runtime is not fully merged, but that
+should not be visible to Go programs.
+</p>
+
+<p>
+The GCC 7 releases are expected to include a complete implementation
+of the Go 1.8 user libraries.  As with earlier releases, the Go 1.8
+runtime is not fully merged, but that should not be visible to Go
+programs.
+</p>
+
 <h2 id="Source_code">Source code</h2>
 
 <p>
@@ -160,23 +173,6 @@ make
 make install
 </pre>
 
-<h3 id="Ubuntu">A note on Ubuntu</h3>
-
-<p>
-Current versions of Ubuntu and versions of GCC before 4.8 disagree on
-where system libraries and header files are found.  This is not a
-gccgo issue.  When building older versions of GCC, setting these
-environment variables while configuring and building gccgo may fix the
-problem.
-</p>
-
-<pre>
-LIBRARY_PATH=/usr/lib/x86_64-linux-gnu
-C_INCLUDE_PATH=/usr/include/x86_64-linux-gnu
-CPLUS_INCLUDE_PATH=/usr/include/x86_64-linux-gnu
-export LIBRARY_PATH C_INCLUDE_PATH CPLUS_INCLUDE_PATH
-</pre>
-
 <h2 id="Using_gccgo">Using gccgo</h2>
 
 <p>
@@ -364,12 +360,15 @@ or with C++ code compiled using <code>extern "C"</code>.
 <h3 id="Types">Types</h3>
 
 <p>
-Basic types map directly: an <code>int</code> in Go is an <code>int</code>
-in C, an <code>int32</code> is an <code>int32_t</code>,
-etc.  Go <code>byte</code> is equivalent to C <code>unsigned
-char</code>.
-Pointers in Go are pointers in C. A Go <code>struct</code> is the same as C
-<code>struct</code> with the same fields and types.
+Basic types map directly: an <code>int32</code> in Go is
+an <code>int32_t</code> in C, an <code>int64</code> is
+an <code>int64_t</code>, etc.
+The Go type <code>int</code> is an integer that is the same size as a
+pointer, and as such corresponds to the C type <code>intptr_t</code>.
+Go <code>byte</code> is equivalent to C <code>unsigned char</code>.
+Pointers in Go are pointers in C.
+A Go <code>struct</code> is the same as C <code>struct</code> with the
+same fields and types.
 </p>
 
 <p>
@@ -380,7 +379,7 @@ structure (this is <b style="color: red;">subject to change</b>):
 <pre>
 struct __go_string {
   const unsigned char *__data;
-  int __length;
+  intptr_t __length;
 };
 </pre>
 
@@ -400,8 +399,8 @@ A slice in Go is a structure.  The current definition is
 <pre>
 struct __go_slice {
   void *__values;
-  int __count;
-  int __capacity;
+  intptr_t __count;
+  intptr_t __capacity;
 };
 </pre>
 
@@ -526,15 +525,3 @@ This procedure is full of unstated caveats and restrictions and we make no
 guarantee that it will not change in the future. It is more useful as a
 starting point for real Go code than as a regular procedure.
 </p>
-
-<h2 id="RTEMS_Port">RTEMS Port</h2>
-<p>
-The gccgo compiler has been ported to <a href="http://www.rtems.com/">
-<code>RTEMS</code></a>. <code>RTEMS</code> is a real-time executive
-that provides a high performance environment for embedded applications
-on a range of processors and embedded hardware. The current gccgo
-port is for x86. The goal is to extend the port to most of the
-<a href="http://www.rtems.org/wiki/index.php/SupportedCPUs">
-architectures supported by <code>RTEMS</code></a>. For more information on the port,
-as well as instructions on how to install it, please see this
-<a href="http://www.rtems.org/wiki/index.php/GCCGoRTEMS"><code>RTEMS</code> Wiki page</a>.

doc/go1.8.html

@@ -15,12 +15,7 @@ Do not send CLs removing the interior tags from such phrases.
 ul li { margin: 0.5em 0; }
 </style>
 
-<h2 id="introduction">DRAFT RELEASE NOTES - Introduction to Go 1.8</h2>
-
-<p><strong>
-Go 1.8 is not yet released. These are work-in-progress
-release notes. Go 1.8 is expected to be released in February 2017.
-</strong></p>
+<h2 id="introduction">Introduction to Go 1.8</h2>
 
 <p>
 The latest Go release, version 1.8, arrives six months after <a href="go1.7">Go 1.7</a>.
@@ -93,7 +88,8 @@ On OpenBSD, Go now requires OpenBSD 5.9 or later. <!-- CL 34093 -->
 <p>
 The Plan 9 port's networking support is now much more complete
 and matches the behavior of Unix and Windows with respect to deadlines
-and cancelation.
+and cancelation. For Plan 9 kernel requirements, see the
+<a href="https://golang.org/wiki/Plan9">Plan 9 wiki page</a>.
 </p>
 
 <p>
@@ -434,11 +430,11 @@ version of gccgo.
 <h3 id="plugin">Plugins</h3>
 
 <p>
-  Go now supports a “<code>plugin</code>” build mode for generating
-  plugins written in Go, and a
+  Go now provides early support for plugins with a “<code>plugin</code>”
+  build mode for generating plugins written in Go, and a
   new <a href="/pkg/plugin/"><code>plugin</code></a> package for
-  loading such plugins at run time. Plugin support is only currently
-  available on Linux.
+  loading such plugins at run time. Plugin support is currently only
+  available on Linux. Please report any issues.
 </p>
 
 <h2 id="runtime">Runtime</h2>
@@ -798,9 +794,9 @@ Optimizations and minor bug fixes are not listed.
       hardware support for AES-GCM is present.
     </p>
 
-    <p> <!-- CL 27315 -->
+    <p> <!-- CL 27315, CL 35290 -->
       AES-128-CBC cipher suites with SHA-256 are also
-      now supported.
+      now supported, but disabled by default.
     </p>
 
   </dd>
@@ -808,11 +804,6 @@ Optimizations and minor bug fixes are not listed.
 
 <dl id="crypto_x509"><dt><a href="/pkg/crypto/x509/">crypto/x509</a></dt>
   <dd>
-    <p> <!-- CL 30578 -->
-      <a href="/pkg/crypto/x509/#SystemCertPool"><code>SystemCertPool</code></a>
-      is now implemented on Windows.
-    </p>
-
     <p> <!-- CL 24743 -->
       PSS signatures are now supported.
     </p>
@@ -863,11 +854,12 @@ crypto/x509: return error for missing SerialNumber (CL 27238)
     <p>
       The <a href="/pkg/database/sql#IsolationLevel"><code>IsolationLevel</code></a>
       can now be set when starting a transaction by setting the isolation level
-      on the <code>Context</code> then passing that <code>Context</code> to
-      <a href="/pkg/database/sql#DB.BeginContext"><code>DB.BeginContext</code></a>.
+      on <a href="/pkg/database/sql#TxOptions.Isolation"><code>TxOptions.Isolation</code></a> and passing
+      it to <a href="/pkg/database/sql#DB.BeginTx"><code>DB.BeginTx</code></a>.
       An error will be returned if an isolation level is selected that the driver
       does not support. A read-only attribute may also be set on the transaction
-      with <a href="/pkg/database/sql/#ReadOnlyContext"><code>ReadOnlyContext</code></a>.
+      by setting <a href="/pkg/database/sql/#TxOptions.ReadOnly"><code>TxOptions.ReadOnly</code></a>
+      to true.
     </p>
     <p>
       Queries now expose the SQL column type information for drivers that support it.
@@ -1312,7 +1304,7 @@ crypto/x509: return error for missing SerialNumber (CL 27238)
         request must have the new
         <a href="/pkg/net/http/#Request"><code>Request.GetBody</code></a>
         field defined.
-        <a href="pkg/net/http/#NewRequest"><code>NewRequest</code></a>
+        <a href="/pkg/net/http/#NewRequest"><code>NewRequest</code></a>
         sets <code>Request.GetBody</code> automatically for common
         body types.
       </li>
@@ -1617,9 +1609,9 @@ crypto/x509: return error for missing SerialNumber (CL 27238)
       June 31 and July 32.
     </p>
 
-    <p> <!-- CL 33029 -->
+    <p> <!-- CL 33029 --> <!-- CL 34816 -->
       The <code>tzdata</code> database has been updated to version
-      2016i for systems that don't already have a local time zone
+      2016j for systems that don't already have a local time zone
       database.
     </p>
 
@@ -1649,6 +1641,17 @@ crypto/x509: return error for missing SerialNumber (CL 27238)
       and only the overall execution of the test binary would fail.
     </p>
 
+    <p><!-- CL 32455 -->
+      The signature of the
+      <a href="/pkg/testing/#MainStart"><code>MainStart</code></a>
+      function has changed, as allowed by the documentation. It is an
+      internal detail and not part of the Go 1 compatibility promise.
+      If you're not calling <code>MainStart</code> directly but see
+      errors, that likely means you set the
+      normally-empty <code>GOROOT</code> environment variable and it
+      doesn't match the version of your <code>go</code> command's binary.
+    </p>
+
   </dd>
 </dl>
 

doc/install-source.html

@@ -147,6 +147,9 @@ either the git branch <code>release-branch.go1.4</code> or
 which contains the Go 1.4 source code plus accumulated fixes
 to keep the tools running on newer operating systems.
 (Go 1.4 was the last distribution in which the tool chain was written in C.)
+After unpacking the Go 1.4 source, <code>cd</code> to
+the <code>src</code> subdirectory and run <code>make.bash</code> (or,
+on Windows, <code>make.bat</code>).
 </p>
 
 <p>
@@ -218,7 +221,7 @@ To build without <code>cgo</code>, set the environment variable
 Change to the directory that will be its parent
 and make sure the <code>go</code> directory does not exist.
 Then clone the repository and check out the latest release tag
-(<code class="versionTag">go1.7.4</code>, for example):</p>
+(<code class="versionTag">go1.8.1</code>, for example):</p>
 
 <pre>
 $ git clone https://go.googlesource.com/go
@@ -406,7 +409,7 @@ New releases are announced on the
 <a href="//groups.google.com/group/golang-announce">golang-announce</a>
 mailing list.
 Each announcement mentions the latest release tag, for instance,
-<code class="versionTag">go1.7.4</code>.
+<code class="versionTag">go1.8.1</code>.
 </p>
 
 <p>

doc/install.html

@@ -250,7 +250,7 @@ $ <b>cd $HOME/go/src/hello</b>
 $ <b>go build</b>
 </pre>
 
-<pre class="testWindows" style="display: none">
+<pre class="testWindows">
 C:\&gt; <b>cd %USERPROFILE%\go\src\hello</b>
 C:\Users\Gopher\go\src\hello&gt; <b>go build</b>
 </pre>
@@ -267,7 +267,7 @@ $ <b>./hello</b>
 hello, world
 </pre>
 
-<pre class="testWindows" style="display: none">
+<pre class="testWindows">
 C:\Users\Gopher\go\src\hello&gt; <b>hello</b>
 hello, world
 </pre>

misc/cgo/test/issue18146.go

@@ -73,7 +73,7 @@ func test18146(t *testing.T) {
 		}
 		runtime.GOMAXPROCS(threads)
 		argv := append(os.Args, "-test.run=NoSuchTestExists")
-		if err := syscall.Exec(os.Args[0], argv, nil); err != nil {
+		if err := syscall.Exec(os.Args[0], argv, os.Environ()); err != nil {
 			t.Fatal(err)
 		}
 	}

misc/cgo/test/sigaltstack.go

@@ -17,7 +17,7 @@ package cgotest
 static stack_t oss;
 static char signalStack[SIGSTKSZ];
 
-static void changeSignalStack() {
+static void changeSignalStack(void) {
 	stack_t ss;
 	memset(&ss, 0, sizeof ss);
 	ss.ss_sp = signalStack;
@@ -29,7 +29,7 @@ static void changeSignalStack() {
 	}
 }
 
-static void restoreSignalStack() {
+static void restoreSignalStack(void) {
 #if (defined(__x86_64__) || defined(__i386__)) && defined(__APPLE__)
 	// The Darwin C library enforces a minimum that the kernel does not.
 	// This is OK since we allocated this much space in mpreinit,
@@ -42,7 +42,7 @@ static void restoreSignalStack() {
 	}
 }
 
-static int zero() {
+static int zero(void) {
 	return 0;
 }
 */

misc/cgo/testcshared/test.bash

@@ -179,6 +179,13 @@ if test "$output" != "PASS"; then
     status=1
 fi
 
+if test "$libext" = "dylib"; then
+	# make sure dylibs are well-formed
+	if ! otool -l libgo*.dylib >/dev/null; then
+		status=1
+	fi
+fi
+
 if test $status = 0; then
     echo "ok"
 fi

misc/cgo/testplugin/src/iface/main.go

@@ -0,0 +1,46 @@
+// Copyright 2017 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package main
+
+import (
+	"iface_i"
+	"log"
+	"plugin"
+)
+
+func main() {
+	a, err := plugin.Open("iface_a.so")
+	if err != nil {
+		log.Fatalf(`plugin.Open("iface_a.so"): %v`, err)
+	}
+	b, err := plugin.Open("iface_b.so")
+	if err != nil {
+		log.Fatalf(`plugin.Open("iface_b.so"): %v`, err)
+	}
+
+	af, err := a.Lookup("F")
+	if err != nil {
+		log.Fatalf(`a.Lookup("F") failed: %v`, err)
+	}
+	bf, err := b.Lookup("F")
+	if err != nil {
+		log.Fatalf(`b.Lookup("F") failed: %v`, err)
+	}
+	if af.(func() interface{})() != bf.(func() interface{})() {
+		panic("empty interfaces not equal")
+	}
+
+	ag, err := a.Lookup("G")
+	if err != nil {
+		log.Fatalf(`a.Lookup("G") failed: %v`, err)
+	}
+	bg, err := b.Lookup("G")
+	if err != nil {
+		log.Fatalf(`b.Lookup("G") failed: %v`, err)
+	}
+	if ag.(func() iface_i.I)() != bg.(func() iface_i.I)() {
+		panic("nonempty interfaces not equal")
+	}
+}

misc/cgo/testplugin/src/iface_a/a.go

@@ -0,0 +1,17 @@
+// Copyright 2017 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package main
+
+import "iface_i"
+
+//go:noinline
+func F() interface{} {
+	return (*iface_i.T)(nil)
+}
+
+//go:noinline
+func G() iface_i.I {
+	return (*iface_i.T)(nil)
+}

misc/cgo/testplugin/src/iface_b/b.go

@@ -0,0 +1,17 @@
+// Copyright 2017 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package main
+
+import "iface_i"
+
+//go:noinline
+func F() interface{} {
+	return (*iface_i.T)(nil)
+}
+
+//go:noinline
+func G() iface_i.I {
+	return (*iface_i.T)(nil)
+}

misc/cgo/testplugin/src/iface_i/i.go

@@ -0,0 +1,17 @@
+// Copyright 2017 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package iface_i
+
+type I interface {
+	M()
+}
+
+type T struct {
+}
+
+func (t *T) M() {
+}
+
+// *T implements I

misc/cgo/testplugin/src/issue18676/dynamodbstreamsevt/definition.go

@@ -0,0 +1,13 @@
+// Copyright 2017 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package dynamodbstreamsevt
+
+import "encoding/json"
+
+var foo json.RawMessage
+
+type Event struct{}
+
+func (e *Event) Dummy() {}

misc/cgo/testplugin/src/issue18676/main.go

@@ -0,0 +1,31 @@
+// Copyright 2017 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// The bug happened like this:
+// 1) The main binary adds an itab for *json.UnsupportedValueError / error
+//    (concrete type / interface type).  This itab goes in hash bucket 0x111.
+// 2) The plugin adds that same itab again.  That makes a cycle in the itab
+//    chain rooted at hash bucket 0x111.
+// 3) The main binary then asks for the itab for *dynamodbstreamsevt.Event /
+//    json.Unmarshaler.  This itab happens to also live in bucket 0x111.
+//    The lookup code goes into an infinite loop searching for this itab.
+// The code is carefully crafted so that the two itabs are both from the
+// same bucket, and so that the second itab doesn't exist in
+// the itab hashmap yet (so the entire linked list must be searched).
+package main
+
+import (
+	"encoding/json"
+	"issue18676/dynamodbstreamsevt"
+	"plugin"
+)
+
+func main() {
+	plugin.Open("plugin.so")
+
+	var x interface{} = (*dynamodbstreamsevt.Event)(nil)
+	if _, ok := x.(json.Unmarshaler); !ok {
+		println("something")
+	}
+}

misc/cgo/testplugin/src/issue18676/plugin.go

@@ -0,0 +1,11 @@
+// Copyright 2017 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package main
+
+import "C"
+
+import "issue18676/dynamodbstreamsevt"
+
+func F(evt *dynamodbstreamsevt.Event) {}

misc/cgo/testplugin/test.bash

@@ -15,8 +15,8 @@ goos=$(go env GOOS)
 goarch=$(go env GOARCH)
 
 function cleanup() {
-	rm -f plugin*.so unnamed*.so
-	rm -rf host pkg sub
+	rm -f plugin*.so unnamed*.so iface*.so
+	rm -rf host pkg sub iface issue18676
 }
 trap cleanup EXIT
 
@@ -32,3 +32,15 @@ GOPATH=$(pwd) go build -buildmode=plugin unnamed2.go
 GOPATH=$(pwd) go build host
 
 LD_LIBRARY_PATH=$(pwd) ./host
+
+# Test that types and itabs get properly uniqified.
+GOPATH=$(pwd) go build -buildmode=plugin iface_a
+GOPATH=$(pwd) go build -buildmode=plugin iface_b
+GOPATH=$(pwd) go build iface
+LD_LIBRARY_PATH=$(pwd) ./iface
+
+# Test for issue 18676 - make sure we don't add the same itab twice.
+# The buggy code hangs forever, so use a timeout to check for that.
+GOPATH=$(pwd) go build -buildmode=plugin -o plugin.so src/issue18676/plugin.go
+GOPATH=$(pwd) go build -o issue18676 src/issue18676/main.go
+timeout 10s ./issue18676

misc/cgo/testsanitizers/msan_shared.go

@@ -0,0 +1,12 @@
+// Copyright 2017 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// This program segfaulted during libpreinit when built with -msan:
+// http://golang.org/issue/18707
+
+package main
+
+import "C"
+
+func main() {}

misc/cgo/testsanitizers/test.bash

@@ -68,6 +68,25 @@ fi
 
 status=0
 
+testmsanshared() {
+  goos=$(go env GOOS)
+  suffix="-installsuffix testsanitizers"
+  libext="so"
+  if [ "$goos" == "darwin" ]; then
+	  libext="dylib"
+  fi
+  go build -msan -buildmode=c-shared $suffix -o ${TMPDIR}/libmsanshared.$libext msan_shared.go
+
+	echo 'int main() { return 0; }' > ${TMPDIR}/testmsanshared.c
+  $CC $(go env GOGCCFLAGS) -fsanitize=memory -o ${TMPDIR}/testmsanshared ${TMPDIR}/testmsanshared.c ${TMPDIR}/libmsanshared.$libext
+
+  if ! LD_LIBRARY_PATH=. ${TMPDIR}/testmsanshared; then
+    echo "FAIL: msan_shared"
+    status=1
+  fi
+  rm -f ${TMPDIR}/{testmsanshared,testmsanshared.c,libmsanshared.$libext}
+}
+
 if test "$msan" = "yes"; then
     if ! go build -msan std; then
 	echo "FAIL: build -msan std"
@@ -108,6 +127,8 @@ if test "$msan" = "yes"; then
 	echo "FAIL: msan_fail"
 	status=1
     fi
+
+    testmsanshared
 fi
 
 if test "$tsan" = "yes"; then

misc/cgo/testshared/shared_test.go

@@ -815,3 +815,14 @@ func TestImplicitInclusion(t *testing.T) {
 	goCmd(t, "install", "-linkshared", "implicitcmd")
 	run(t, "running executable linked against library that contains same package as it", "./bin/implicitcmd")
 }
+
+// Tests to make sure that the type fields of empty interfaces and itab
+// fields of nonempty interfaces are unique even across modules,
+// so that interface equality works correctly.
+func TestInterface(t *testing.T) {
+	goCmd(t, "install", "-buildmode=shared", "-linkshared", "iface_a")
+	// Note: iface_i gets installed implicitly as a dependency of iface_a.
+	goCmd(t, "install", "-buildmode=shared", "-linkshared", "iface_b")
+	goCmd(t, "install", "-linkshared", "iface")
+	run(t, "running type/itab uniqueness tester", "./bin/iface")
+}

misc/cgo/testshared/src/depBase/dep.go

@@ -5,6 +5,8 @@ import (
 	"reflect"
 )
 
+var SlicePtr interface{} = &[]int{}
+
 var V int = 1
 
 var HasMask []string = []string{"hi"}

misc/cgo/testshared/src/exe/exe.go

@@ -19,6 +19,8 @@ func F() *C {
 	return nil
 }
 
+var slicePtr interface{} = &[]int{}
+
 func main() {
 	defer depBase.ImplementedInAsm()
 	// This code below causes various go.itab.* symbols to be generated in
@@ -32,4 +34,11 @@ func main() {
 	if reflect.TypeOf(F).Out(0) != reflect.TypeOf(c) {
 		panic("bad reflection results, see golang.org/issue/18252")
 	}
+
+	sp := reflect.New(reflect.TypeOf(slicePtr).Elem())
+	s := sp.Interface()
+
+	if reflect.TypeOf(s) != reflect.TypeOf(slicePtr) {
+		panic("bad reflection results, see golang.org/issue/18729")
+	}
 }

misc/cgo/testshared/src/iface/main.go

@@ -0,0 +1,17 @@
+// Copyright 2017 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package main
+
+import "iface_a"
+import "iface_b"
+
+func main() {
+	if iface_a.F() != iface_b.F() {
+		panic("empty interfaces not equal")
+	}
+	if iface_a.G() != iface_b.G() {
+		panic("non-empty interfaces not equal")
+	}
+}

misc/cgo/testshared/src/iface_a/a.go

@@ -0,0 +1,17 @@
+// Copyright 2017 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package iface_a
+
+import "iface_i"
+
+//go:noinline
+func F() interface{} {
+	return (*iface_i.T)(nil)
+}
+
+//go:noinline
+func G() iface_i.I {
+	return (*iface_i.T)(nil)
+}

misc/cgo/testshared/src/iface_b/b.go

@@ -0,0 +1,17 @@
+// Copyright 2017 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package iface_b
+
+import "iface_i"
+
+//go:noinline
+func F() interface{} {
+	return (*iface_i.T)(nil)
+}
+
+//go:noinline
+func G() iface_i.I {
+	return (*iface_i.T)(nil)
+}

misc/cgo/testshared/src/iface_i/i.go

@@ -0,0 +1,17 @@
+// Copyright 2017 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package iface_i
+
+type I interface {
+	M()
+}
+
+type T struct {
+}
+
+func (t *T) M() {
+}
+
+// *T implements I

misc/ios/go_darwin_arm_exec.go

@@ -99,7 +99,7 @@ func main() {
 	// Approximately 1 in a 100 binaries fail to start. If it happens,
 	// try again. These failures happen for several reasons beyond
 	// our control, but all of them are safe to retry as they happen
-	// before lldb encounters the initial getwd breakpoint. As we
+	// before lldb encounters the initial SIGUSR2 stop. As we
 	// know the tests haven't started, we are not hiding flaky tests
 	// with this retry.
 	for i := 0; i < 5; i++ {
@@ -204,6 +204,11 @@ func run(bin string, args []string) (err error) {
 	var opts options
 	opts, args = parseArgs(args)
 
+	// Pass the suffix for the current working directory as the
+	// first argument to the test. For iOS, cmd/go generates
+	// special handling of this argument.
+	args = append([]string{"cwdSuffix=" + pkgpath}, args...)
+
 	// ios-deploy invokes lldb to give us a shell session with the app.
 	s, err := newSession(appdir, args, opts)
 	if err != nil {
@@ -224,6 +229,7 @@ func run(bin string, args []string) (err error) {
 	s.do(`process handle SIGHUP  --stop false --pass true --notify false`)
 	s.do(`process handle SIGPIPE --stop false --pass true --notify false`)
 	s.do(`process handle SIGUSR1 --stop false --pass true --notify false`)
+	s.do(`process handle SIGUSR2 --stop true --pass false --notify true`) // sent by test harness
 	s.do(`process handle SIGCONT --stop false --pass true --notify false`)
 	s.do(`process handle SIGSEGV --stop false --pass true --notify false`) // does not work
 	s.do(`process handle SIGBUS  --stop false --pass true --notify false`) // does not work
@@ -236,20 +242,9 @@ func run(bin string, args []string) (err error) {
 		return nil
 	}
 
-	s.do(`breakpoint set -n getwd`) // in runtime/cgo/gcc_darwin_arm.go
-
 	started = true
 
-	s.doCmd("run", "stop reason = breakpoint", 20*time.Second)
-
-	// Move the current working directory into the faux gopath.
-	if pkgpath != "src" {
-		s.do(`breakpoint delete 1`)
-		s.do(`expr char* $mem = (char*)malloc(512)`)
-		s.do(`expr $mem = (char*)getwd($mem, 512)`)
-		s.do(`expr $mem = (char*)strcat($mem, "/` + pkgpath + `")`)
-		s.do(`call (void)chdir($mem)`)
-	}
+	s.doCmd("run", "stop reason = signal SIGUSR2", 20*time.Second)
 
 	startTestsLen := s.out.Len()
 	fmt.Fprintln(s.in, `process continue`)
@@ -520,13 +515,11 @@ func copyLocalData(dstbase string) (pkgpath string, err error) {
 
 	// Copy timezone file.
 	//
-	// Typical apps have the zoneinfo.zip in the root of their app bundle,
+	// Apps have the zoneinfo.zip in the root of their app bundle,
 	// read by the time package as the working directory at initialization.
-	// As we move the working directory to the GOROOT pkg directory, we
-	// install the zoneinfo.zip file in the pkgpath.
 	if underGoRoot {
 		err := cp(
-			filepath.Join(dstbase, pkgpath),
+			dstbase,
 			filepath.Join(cwd, "lib", "time", "zoneinfo.zip"),
 		)
 		if err != nil {

src/cmd/compile/internal/amd64/galign.go

@@ -27,4 +27,5 @@ func Init() {
 	gc.Thearch.SSAMarkMoves = ssaMarkMoves
 	gc.Thearch.SSAGenValue = ssaGenValue
 	gc.Thearch.SSAGenBlock = ssaGenBlock
+	gc.Thearch.ZeroAuto = zeroAuto
 }

src/cmd/compile/internal/amd64/ggen.go

@@ -166,6 +166,27 @@ func zerorange(p *obj.Prog, frame int64, lo int64, hi int64, ax *uint32, x0 *uin
 	return p
 }
 
+func zeroAuto(n *gc.Node, pp *obj.Prog) {
+	// Note: this code must not clobber any registers.
+	op := x86.AMOVQ
+	if gc.Widthptr == 4 {
+		op = x86.AMOVL
+	}
+	sym := gc.Linksym(n.Sym)
+	size := n.Type.Size()
+	for i := int64(0); i < size; i += int64(gc.Widthptr) {
+		p := gc.AddAsmAfter(op, pp)
+		pp = p
+		p.From.Type = obj.TYPE_CONST
+		p.From.Offset = 0
+		p.To.Type = obj.TYPE_MEM
+		p.To.Name = obj.NAME_AUTO
+		p.To.Reg = x86.REG_SP
+		p.To.Offset = n.Xoffset + i
+		p.To.Sym = sym
+	}
+}
+
 func ginsnop() {
 	// This is actually not the x86 NOP anymore,
 	// but at the point where it gets used, AX is dead

src/cmd/compile/internal/arm/galign.go

@@ -21,4 +21,5 @@ func Init() {
 	gc.Thearch.SSAMarkMoves = func(s *gc.SSAGenState, b *ssa.Block) {}
 	gc.Thearch.SSAGenValue = ssaGenValue
 	gc.Thearch.SSAGenBlock = ssaGenBlock
+	gc.Thearch.ZeroAuto = zeroAuto
 }

src/cmd/compile/internal/arm/ggen.go

@@ -92,6 +92,27 @@ func zerorange(p *obj.Prog, frame int64, lo int64, hi int64, r0 *uint32) *obj.Pr
 
 	return p
 }
+func zeroAuto(n *gc.Node, pp *obj.Prog) {
+	// Note: this code must not clobber any registers.
+	sym := gc.Linksym(n.Sym)
+	size := n.Type.Size()
+	p := gc.Prog(arm.AMOVW)
+	p.From.Type = obj.TYPE_CONST
+	p.From.Offset = 0
+	p.To.Type = obj.TYPE_REG
+	p.To.Reg = arm.REGTMP
+	for i := int64(0); i < size; i += 4 {
+		p := gc.AddAsmAfter(arm.AMOVW, pp)
+		pp = p
+		p.From.Type = obj.TYPE_REG
+		p.From.Reg = arm.REGTMP
+		p.To.Type = obj.TYPE_MEM
+		p.To.Name = obj.NAME_AUTO
+		p.To.Reg = arm.REGSP
+		p.To.Offset = n.Xoffset + i
+		p.To.Sym = sym
+	}
+}
 
 func ginsnop() {
 	p := gc.Prog(arm.AAND)

src/cmd/compile/internal/arm64/galign.go

@@ -21,4 +21,5 @@ func Init() {
 	gc.Thearch.SSAMarkMoves = func(s *gc.SSAGenState, b *ssa.Block) {}
 	gc.Thearch.SSAGenValue = ssaGenValue
 	gc.Thearch.SSAGenBlock = ssaGenBlock
+	gc.Thearch.ZeroAuto = zeroAuto
 }

src/cmd/compile/internal/arm64/ggen.go

@@ -103,6 +103,23 @@ func zerorange(p *obj.Prog, frame int64, lo int64, hi int64) *obj.Prog {
 	return p
 }
 
+func zeroAuto(n *gc.Node, pp *obj.Prog) {
+	// Note: this code must not clobber any registers.
+	sym := gc.Linksym(n.Sym)
+	size := n.Type.Size()
+	for i := int64(0); i < size; i += 8 {
+		p := gc.AddAsmAfter(arm64.AMOVD, pp)
+		pp = p
+		p.From.Type = obj.TYPE_REG
+		p.From.Reg = arm64.REGZERO
+		p.To.Type = obj.TYPE_MEM
+		p.To.Name = obj.NAME_AUTO
+		p.To.Reg = arm64.REGSP
+		p.To.Offset = n.Xoffset + i
+		p.To.Sym = sym
+	}
+}
+
 func ginsnop() {
 	p := gc.Prog(arm64.AHINT)
 	p.From.Type = obj.TYPE_CONST

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

@@ -303,7 +303,9 @@ func genhash(sym *Sym, t *Type) {
 	typecheckslice(fn.Nbody.Slice(), Etop)
 	Curfn = nil
 	popdcl()
-	testdclstack()
+	if debug_dclstack != 0 {
+		testdclstack()
+	}
 
 	// Disable safemode while compiling this code: the code we
 	// generate internally can refer to unsafe.Pointer.
@@ -493,7 +495,9 @@ func geneq(sym *Sym, t *Type) {
 	typecheckslice(fn.Nbody.Slice(), Etop)
 	Curfn = nil
 	popdcl()
-	testdclstack()
+	if debug_dclstack != 0 {
+		testdclstack()
+	}
 
 	// Disable safemode while compiling this code: the code we
 	// generate internally can refer to unsafe.Pointer.

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

@@ -217,7 +217,9 @@ func Import(in *bufio.Reader) {
 	typecheckok = tcok
 	resumecheckwidth()
 
-	testdclstack() // debugging only
+	if debug_dclstack != 0 {
+		testdclstack()
+	}
 }
 
 func formatErrorf(format string, args ...interface{}) {

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

@@ -15,6 +15,7 @@ var runtimeDecls = [...]struct {
 	{"panicwrap", funcTag, 7},
 	{"gopanic", funcTag, 9},
 	{"gorecover", funcTag, 12},
+	{"goschedguarded", funcTag, 5},
 	{"printbool", funcTag, 14},
 	{"printfloat", funcTag, 16},
 	{"printint", funcTag, 18},

src/cmd/compile/internal/gc/builtin/runtime.go

@@ -21,6 +21,7 @@ func panicwrap(string, string, string)
 
 func gopanic(interface{})
 func gorecover(*int32) interface{}
+func goschedguarded()
 
 func printbool(bool)
 func printfloat(float64)

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

@@ -477,6 +477,10 @@ func escAnalyze(all []*Node, recursive bool) {
 	for _, n := range all {
 		if n.Op == ODCLFUNC {
 			n.Esc = EscFuncPlanned
+			if Debug['m'] > 3 {
+				Dump("escAnalyze", n)
+			}
+
 		}
 	}
 
@@ -1675,7 +1679,10 @@ func (e *EscState) escflows(dst, src *Node, why *EscStep) {
 	}
 
 	// Don't bother building a graph for scalars.
-	if src.Type != nil && !haspointers(src.Type) {
+	if src.Type != nil && !haspointers(src.Type) && !isReflectHeaderDataField(src) {
+		if Debug['m'] > 3 {
+			fmt.Printf("%v::NOT flows:: %S <- %S\n", linestr(lineno), dst, src)
+		}
 		return
 	}
 

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

@@ -364,6 +364,12 @@ type Arch struct {
 	// SSAGenBlock emits end-of-block Progs. SSAGenValue should be called
 	// for all values in the block before SSAGenBlock.
 	SSAGenBlock func(s *SSAGenState, b, next *ssa.Block)
+
+	// ZeroAuto emits code to zero the given auto stack variable.
+	// Code is added immediately after pp.
+	// ZeroAuto must not use any non-temporary registers.
+	// ZeroAuto will only be called for variables which contain a pointer.
+	ZeroAuto func(n *Node, pp *obj.Prog)
 }
 
 var pcloc int32

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

@@ -72,6 +72,15 @@ func Appendpp(p *obj.Prog, as obj.As, ftype obj.AddrType, freg int16, foffset in
 	return q
 }
 
+func AddAsmAfter(as obj.As, p *obj.Prog) *obj.Prog {
+	q := Ctxt.NewProg()
+	Clearp(q)
+	q.As = as
+	q.Link = p.Link
+	p.Link = q
+	return q
+}
+
 func ggloblnod(nam *Node) {
 	s := Linksym(nam.Sym)
 	s.Gotype = Linksym(ngotype(nam))

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

@@ -30,11 +30,12 @@ var (
 )
 
 var (
-	Debug_append  int
-	Debug_closure int
-	Debug_panic   int
-	Debug_slice   int
-	Debug_wb      int
+	Debug_append   int
+	Debug_closure  int
+	debug_dclstack int
+	Debug_panic    int
+	Debug_slice    int
+	Debug_wb       int
 )
 
 // Debug arguments.
@@ -48,6 +49,7 @@ var debugtab = []struct {
 	{"append", &Debug_append},         // print information about append compilation
 	{"closure", &Debug_closure},       // print information about closure compilation
 	{"disablenil", &disable_checknil}, // disable nil checks
+	{"dclstack", &debug_dclstack},     // run internal dclstack checks
 	{"gcprog", &Debug_gcprog},         // print dump of GC programs
 	{"nil", &Debug_checknil},          // print information about nil checks
 	{"panic", &Debug_panic},           // do not hide any compiler panic
@@ -325,7 +327,6 @@ func Main() {
 	timings.Stop()
 	timings.AddEvent(int64(lexlineno-lexlineno0), "lines")
 
-	testdclstack()
 	mkpackage(localpkg.Name) // final import not used checks
 	finishUniverse()
 

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

@@ -34,6 +34,7 @@ func parseFile(filename string) {
 	}
 
 	if nsyntaxerrors == 0 {
+		// Always run testdclstack here, even when debug_dclstack is not set, as a sanity measure.
 		testdclstack()
 	}
 }
@@ -1051,6 +1052,7 @@ func (p *noder) pragma(pos, line int, text string) syntax.Pragma {
 		lookup(f[1]).Linkname = f[2]
 
 	case strings.HasPrefix(text, "go:cgo_"):
+		lineno = p.baseline + int32(line) - 1 // pragcgo may call yyerror
 		pragcgobuf += pragcgo(text)
 		fallthrough // because of //go:cgo_unsafe_args
 	default:
@@ -1058,6 +1060,7 @@ func (p *noder) pragma(pos, line int, text string) syntax.Pragma {
 		if i := strings.Index(text, " "); i >= 0 {
 			verb = verb[:i]
 		}
+		lineno = p.baseline + int32(line) - 1 // pragmaValue may call yyerror
 		return syntax.Pragma(pragmaValue(verb))
 	}
 

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

@@ -120,7 +120,30 @@ func Gvarlive(n *Node) {
 }
 
 func removevardef(firstp *obj.Prog) {
+	// At VARKILLs, zero variable if it is ambiguously live.
+	// After the VARKILL anything this variable references
+	// might be collected. If it were to become live again later,
+	// the GC will see references to already-collected objects.
+	// See issue 20029.
 	for p := firstp; p != nil; p = p.Link {
+		if p.As != obj.AVARKILL {
+			continue
+		}
+		n := p.To.Node.(*Node)
+		if !n.Name.Needzero {
+			continue
+		}
+		if n.Class != PAUTO {
+			Fatalf("zero of variable which isn't PAUTO %v", n)
+		}
+		if n.Type.Size()%int64(Widthptr) != 0 {
+			Fatalf("zero of variable not a multiple of ptr size %v", n)
+		}
+		Thearch.ZeroAuto(n, p)
+	}
+
+	for p := firstp; p != nil; p = p.Link {
+
 		for p.Link != nil && (p.Link.As == obj.AVARDEF || p.Link.As == obj.AVARKILL || p.Link.As == obj.AVARLIVE) {
 			p.Link = p.Link.Link
 		}

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

@@ -585,7 +585,7 @@ func isliteral(n *Node) bool {
 }
 
 func (n *Node) isSimpleName() bool {
-	return n.Op == ONAME && n.Addable && n.Class != PAUTOHEAP
+	return n.Op == ONAME && n.Addable && n.Class != PAUTOHEAP && n.Class != PEXTERN
 }
 
 func litas(l *Node, r *Node, init *Nodes) {

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

@@ -64,6 +64,9 @@ func buildssa(fn *Node) *ssa.Func {
 	s.config = initssa()
 	s.f = s.config.NewFunc()
 	s.f.Name = name
+	if fn.Func.Pragma&Nosplit != 0 {
+		s.f.NoSplit = true
+	}
 	s.exitCode = fn.Func.Exit
 	s.panics = map[funcLine]*ssa.Block{}
 	s.config.DebugTest = s.config.DebugHashMatch("GOSSAHASH", name)
@@ -3467,8 +3470,13 @@ func (s *state) insertWBstore(t *Type, left, right *ssa.Value, line int32, skip
 	if s.WBLineno == 0 {
 		s.WBLineno = left.Line
 	}
-	s.storeTypeScalars(t, left, right, skip)
-	s.storeTypePtrsWB(t, left, right)
+	if t == Types[TUINTPTR] {
+		// Stores to reflect.{Slice,String}Header.Data.
+		s.vars[&memVar] = s.newValue3I(ssa.OpStoreWB, ssa.TypeMem, s.config.PtrSize, left, right, s.mem())
+	} else {
+		s.storeTypeScalars(t, left, right, skip)
+		s.storeTypePtrsWB(t, left, right)
+	}
 
 	// WB ops will be expanded to branches at writebarrier phase.
 	// To make it easy, we put WB ops at the end of a block, so

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

@@ -63,6 +63,9 @@ func TestArithmeticConst(t *testing.T) { runTest(t, "arithConst.go") }
 
 func TestChan(t *testing.T) { runTest(t, "chan.go") }
 
+// TestComparisonsConst tests results for comparison operations against constants.
+func TestComparisonsConst(t *testing.T) { runTest(t, "cmpConst.go") }
+
 func TestCompound(t *testing.T) { runTest(t, "compound.go") }
 
 func TestCtl(t *testing.T) { runTest(t, "ctl.go") }

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

@@ -1833,7 +1833,9 @@ func genwrapper(rcvr *Type, method *Field, newnam *Sym, iface int) {
 	funcbody(fn)
 	Curfn = fn
 	popdcl()
-	testdclstack()
+	if debug_dclstack != 0 {
+		testdclstack()
+	}
 
 	// wrappers where T is anonymous (struct or interface) can be duplicated.
 	if rcvr.IsStruct() || rcvr.IsInterface() || rcvr.IsPtr() && rcvr.Elem().IsStruct() {

src/cmd/compile/internal/gc/testdata/arith.go

@@ -488,6 +488,17 @@ func testLrot() {
 			wantA, wantB, wantC, wantD, ", got", a, b, c, d)
 		failed = true
 	}
+	// Also test inputs with the top bit set, and make sure
+	// sub-word right shift has high bits cleared first.
+	// See issue #19270.
+	wantA, wantB, wantC, wantD = uint8(0xdf), uint16(0xdfff),
+		uint32(0xdfffffff), uint64(0xdfffffffffffffff)
+	a, b, c, d = lrot1_ssa(0xfe, 0xfffe, 0xfffffffe, 0xfffffffffffffffe)
+	if a != wantA || b != wantB || c != wantC || d != wantD {
+		println("lrot1_ssa(0xfe, 0xfffe, 0xfffffffe, 0xfffffffffffffffe)=",
+			wantA, wantB, wantC, wantD, ", got", a, b, c, d)
+		failed = true
+	}
 	x := lrot2_ssa(0xb0000001, 32)
 	wantX := uint32(0xb0000001)
 	if x != wantX {

src/cmd/compile/internal/gc/testdata/gen/cmpConstGen.go

@@ -0,0 +1,248 @@
+// Copyright 2017 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// This program generates a test to verify that the standard comparison
+// operators properly handle one const operand. The test file should be
+// generated with a known working version of go.
+// launch with `go run cmpConstGen.go` a file called cmpConst.go
+// will be written into the parent directory containing the tests
+
+package main
+
+import (
+	"bytes"
+	"fmt"
+	"go/format"
+	"io/ioutil"
+	"log"
+	"math/big"
+	"sort"
+)
+
+const (
+	maxU64 = (1 << 64) - 1
+	maxU32 = (1 << 32) - 1
+	maxU16 = (1 << 16) - 1
+	maxU8  = (1 << 8) - 1
+
+	maxI64 = (1 << 63) - 1
+	maxI32 = (1 << 31) - 1
+	maxI16 = (1 << 15) - 1
+	maxI8  = (1 << 7) - 1
+
+	minI64 = -(1 << 63)
+	minI32 = -(1 << 31)
+	minI16 = -(1 << 15)
+	minI8  = -(1 << 7)
+)
+
+func cmp(left *big.Int, op string, right *big.Int) bool {
+	switch left.Cmp(right) {
+	case -1: // less than
+		return op == "<" || op == "<=" || op == "!="
+	case 0: // equal
+		return op == "==" || op == "<=" || op == ">="
+	case 1: // greater than
+		return op == ">" || op == ">=" || op == "!="
+	}
+	panic("unexpected comparison value")
+}
+
+func inRange(typ string, val *big.Int) bool {
+	min, max := &big.Int{}, &big.Int{}
+	switch typ {
+	case "uint64":
+		max = max.SetUint64(maxU64)
+	case "uint32":
+		max = max.SetUint64(maxU32)
+	case "uint16":
+		max = max.SetUint64(maxU16)
+	case "uint8":
+		max = max.SetUint64(maxU8)
+	case "int64":
+		min = min.SetInt64(minI64)
+		max = max.SetInt64(maxI64)
+	case "int32":
+		min = min.SetInt64(minI32)
+		max = max.SetInt64(maxI32)
+	case "int16":
+		min = min.SetInt64(minI16)
+		max = max.SetInt64(maxI16)
+	case "int8":
+		min = min.SetInt64(minI8)
+		max = max.SetInt64(maxI8)
+	default:
+		panic("unexpected type")
+	}
+	return cmp(min, "<=", val) && cmp(val, "<=", max)
+}
+
+func getValues(typ string) []*big.Int {
+	Uint := func(v uint64) *big.Int { return big.NewInt(0).SetUint64(v) }
+	Int := func(v int64) *big.Int { return big.NewInt(0).SetInt64(v) }
+	values := []*big.Int{
+		// limits
+		Uint(maxU64),
+		Uint(maxU64 - 1),
+		Uint(maxI64 + 1),
+		Uint(maxI64),
+		Uint(maxI64 - 1),
+		Uint(maxU32 + 1),
+		Uint(maxU32),
+		Uint(maxU32 - 1),
+		Uint(maxI32 + 1),
+		Uint(maxI32),
+		Uint(maxI32 - 1),
+		Uint(maxU16 + 1),
+		Uint(maxU16),
+		Uint(maxU16 - 1),
+		Uint(maxI16 + 1),
+		Uint(maxI16),
+		Uint(maxI16 - 1),
+		Uint(maxU8 + 1),
+		Uint(maxU8),
+		Uint(maxU8 - 1),
+		Uint(maxI8 + 1),
+		Uint(maxI8),
+		Uint(maxI8 - 1),
+		Uint(0),
+		Int(minI8 + 1),
+		Int(minI8),
+		Int(minI8 - 1),
+		Int(minI16 + 1),
+		Int(minI16),
+		Int(minI16 - 1),
+		Int(minI32 + 1),
+		Int(minI32),
+		Int(minI32 - 1),
+		Int(minI64 + 1),
+		Int(minI64),
+
+		// other possibly interesting values
+		Uint(1),
+		Int(-1),
+		Uint(0xff << 56),
+		Uint(0xff << 32),
+		Uint(0xff << 24),
+	}
+	sort.Slice(values, func(i, j int) bool { return values[i].Cmp(values[j]) == -1 })
+	var ret []*big.Int
+	for _, val := range values {
+		if !inRange(typ, val) {
+			continue
+		}
+		ret = append(ret, val)
+	}
+	return ret
+}
+
+func sigString(v *big.Int) string {
+	var t big.Int
+	t.Abs(v)
+	if v.Sign() == -1 {
+		return "neg" + t.String()
+	}
+	return t.String()
+}
+
+func main() {
+	types := []string{
+		"uint64", "uint32", "uint16", "uint8",
+		"int64", "int32", "int16", "int8",
+	}
+
+	w := new(bytes.Buffer)
+	fmt.Fprintf(w, "// run\n")
+	fmt.Fprintf(w, "// Code generated by gen/cmpConstGen.go. DO NOT EDIT.\n\n")
+	fmt.Fprintf(w, "package main;\n")
+	fmt.Fprintf(w, "import (\"fmt\"; \"reflect\"; \"runtime\";)\n")
+	fmt.Fprintf(w, "// results show the expected result for the elements left of, equal to and right of the index.\n")
+	fmt.Fprintf(w, "type result struct{l, e, r bool}\n")
+	fmt.Fprintf(w, "var (\n")
+	fmt.Fprintf(w, "	eq = result{l: false, e: true, r: false}\n")
+	fmt.Fprintf(w, "	ne = result{l: true, e: false, r: true}\n")
+	fmt.Fprintf(w, "	lt = result{l: true, e: false, r: false}\n")
+	fmt.Fprintf(w, "	le = result{l: true, e: true, r: false}\n")
+	fmt.Fprintf(w, "	gt = result{l: false, e: false, r: true}\n")
+	fmt.Fprintf(w, "	ge = result{l: false, e: true, r: true}\n")
+	fmt.Fprintf(w, ")\n")
+
+	operators := []struct{ op, name string }{
+		{"<", "lt"},
+		{"<=", "le"},
+		{">", "gt"},
+		{">=", "ge"},
+		{"==", "eq"},
+		{"!=", "ne"},
+	}
+
+	for _, typ := range types {
+		// generate a slice containing valid values for this type
+		fmt.Fprintf(w, "\n// %v tests\n", typ)
+		values := getValues(typ)
+		fmt.Fprintf(w, "var %v_vals = []%v{\n", typ, typ)
+		for _, val := range values {
+			fmt.Fprintf(w, "%v,\n", val.String())
+		}
+		fmt.Fprintf(w, "}\n")
+
+		// generate test functions
+		for _, r := range values {
+			// TODO: could also test constant on lhs.
+			sig := sigString(r)
+			for _, op := range operators {
+				// no need for go:noinline because the function is called indirectly
+				fmt.Fprintf(w, "func %v_%v_%v(x %v) bool { return x %v %v; }\n", op.name, sig, typ, typ, op.op, r.String())
+			}
+		}
+
+		// generate a table of test cases
+		fmt.Fprintf(w, "var %v_tests = []struct{\n", typ)
+		fmt.Fprintf(w, "	idx int // index of the constant used\n")
+		fmt.Fprintf(w, "	exp result // expected results\n")
+		fmt.Fprintf(w, "	fn  func(%v) bool\n", typ)
+		fmt.Fprintf(w, "}{\n")
+		for i, r := range values {
+			sig := sigString(r)
+			for _, op := range operators {
+				fmt.Fprintf(w, "{idx: %v,", i)
+				fmt.Fprintf(w, "exp: %v,", op.name)
+				fmt.Fprintf(w, "fn:  %v_%v_%v},\n", op.name, sig, typ)
+			}
+		}
+		fmt.Fprintf(w, "}\n")
+	}
+
+	// emit the main function, looping over all test cases
+	fmt.Fprintf(w, "func main() {\n")
+	for _, typ := range types {
+		fmt.Fprintf(w, "for i, test := range %v_tests {\n", typ)
+		fmt.Fprintf(w, "	for j, x := range %v_vals {\n", typ)
+		fmt.Fprintf(w, "		want := test.exp.l\n")
+		fmt.Fprintf(w, "		if j == test.idx {\nwant = test.exp.e\n}")
+		fmt.Fprintf(w, "		else if j > test.idx {\nwant = test.exp.r\n}\n")
+		fmt.Fprintf(w, "		if test.fn(x) != want {\n")
+		fmt.Fprintf(w, "			fn := runtime.FuncForPC(reflect.ValueOf(test.fn).Pointer()).Name()\n")
+		fmt.Fprintf(w, "			msg := fmt.Sprintf(\"test failed: %%v(%%v) != %%v [type=%v i=%%v j=%%v idx=%%v]\", fn, x, want, i, j, test.idx)\n", typ)
+		fmt.Fprintf(w, "			panic(msg)\n")
+		fmt.Fprintf(w, "		}\n")
+		fmt.Fprintf(w, "	}\n")
+		fmt.Fprintf(w, "}\n")
+	}
+	fmt.Fprintf(w, "}\n")
+
+	// gofmt result
+	b := w.Bytes()
+	src, err := format.Source(b)
+	if err != nil {
+		fmt.Printf("%s\n", b)
+		panic(err)
+	}
+
+	// write to file
+	err = ioutil.WriteFile("../cmpConst.go", src, 0666)
+	if err != nil {
+		log.Fatalf("can't write output: %v\n", err)
+	}
+}

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

@@ -1114,7 +1114,7 @@ OpSwitch:
 
 	case OSLICE, OSLICE3:
 		ok |= Erv
-		n.Left = typecheck(n.Left, top)
+		n.Left = typecheck(n.Left, Erv)
 		low, high, max := n.SliceBounds()
 		hasmax := n.Op.IsSlice3()
 		low = typecheck(low, Erv)
@@ -1126,6 +1126,10 @@ OpSwitch:
 		max = indexlit(max)
 		n.SetSliceBounds(low, high, max)
 		l := n.Left
+		if l.Type == nil {
+			n.Type = nil
+			return n
+		}
 		if l.Type.IsArray() {
 			if !islvalue(n.Left) {
 				yyerror("invalid operation %v (slice of unaddressable value)", n)
@@ -1138,12 +1142,7 @@ OpSwitch:
 			n.Left = typecheck(n.Left, Erv)
 			l = n.Left
 		}
-
 		t := l.Type
-		if t == nil {
-			n.Type = nil
-			return n
-		}
 		var tp *Type
 		if t.IsString() {
 			if hasmax {

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

@@ -57,8 +57,13 @@ func startProfile() {
 			Fatalf("%v", err)
 		}
 		atExit(func() {
-			runtime.GC() // profile all outstanding allocations
-			if err := pprof.WriteHeapProfile(f); err != nil {
+			// Profile all outstanding allocations.
+			runtime.GC()
+			// compilebench parses the memory profile to extract memstats,
+			// which are only written in the legacy pprof format.
+			// See golang.org/issue/18641 and runtime/pprof/pprof.go:writeHeap.
+			const writeLegacyFormat = 1
+			if err := pprof.Lookup("heap").WriteTo(f, writeLegacyFormat); err != nil {
 				Fatalf("%v", err)
 			}
 		})

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

@@ -2071,6 +2071,29 @@ func isstack(n *Node) bool {
 	return false
 }
 
+// isReflectHeaderDataField reports whether l is an expression p.Data
+// where p has type reflect.SliceHeader or reflect.StringHeader.
+func isReflectHeaderDataField(l *Node) bool {
+	if l.Type != Types[TUINTPTR] {
+		return false
+	}
+
+	var tsym *Sym
+	switch l.Op {
+	case ODOT:
+		tsym = l.Left.Type.Sym
+	case ODOTPTR:
+		tsym = l.Left.Type.Elem().Sym
+	default:
+		return false
+	}
+
+	if tsym == nil || l.Sym.Name != "Data" || tsym.Pkg.Path != "reflect" {
+		return false
+	}
+	return tsym.Name == "SliceHeader" || tsym.Name == "StringHeader"
+}
+
 // Do we need a write barrier for the assignment l = r?
 func needwritebarrier(l *Node, r *Node) bool {
 	if !use_writebarrier {
@@ -2081,15 +2104,21 @@ func needwritebarrier(l *Node, r *Node) bool {
 		return false
 	}
 
-	// No write barrier for write of non-pointers.
-	dowidth(l.Type)
-
-	if !haspointers(l.Type) {
+	// No write barrier for write to stack.
+	if isstack(l) {
 		return false
 	}
 
-	// No write barrier for write to stack.
-	if isstack(l) {
+	// Package unsafe's documentation says storing pointers into
+	// reflect.SliceHeader and reflect.StringHeader's Data fields
+	// is valid, even though they have type uintptr (#19168).
+	if isReflectHeaderDataField(l) {
+		return true
+	}
+
+	// No write barrier for write of non-pointers.
+	dowidth(l.Type)
+	if !haspointers(l.Type) {
 		return false
 	}
 
@@ -3117,12 +3146,12 @@ func walkcompare(n *Node, init *Nodes) *Node {
 		cmpr = cmpr.Left
 	}
 
-	if !islvalue(cmpl) || !islvalue(cmpr) {
-		Fatalf("arguments of comparison must be lvalues - %v %v", cmpl, cmpr)
-	}
-
 	// Chose not to inline. Call equality function directly.
 	if !inline {
+		if !islvalue(cmpl) || !islvalue(cmpr) {
+			Fatalf("arguments of comparison must be lvalues - %v %v", cmpl, cmpr)
+		}
+
 		// eq algs take pointers
 		pl := temp(ptrto(t))
 		al := nod(OAS, pl, nod(OADDR, cmpl, nil))

src/cmd/compile/internal/mips/galign.go

@@ -23,4 +23,5 @@ func Init() {
 	gc.Thearch.SSAMarkMoves = func(s *gc.SSAGenState, b *ssa.Block) {}
 	gc.Thearch.SSAGenValue = ssaGenValue
 	gc.Thearch.SSAGenBlock = ssaGenBlock
+	gc.Thearch.ZeroAuto = zeroAuto
 }

src/cmd/compile/internal/mips/ggen.go

@@ -92,6 +92,23 @@ func zerorange(p *obj.Prog, frame int64, lo int64, hi int64) *obj.Prog {
 	return p
 }
 
+func zeroAuto(n *gc.Node, pp *obj.Prog) {
+	// Note: this code must not clobber any registers.
+	sym := gc.Linksym(n.Sym)
+	size := n.Type.Size()
+	for i := int64(0); i < size; i += 4 {
+		p := gc.AddAsmAfter(mips.AMOVW, pp)
+		pp = p
+		p.From.Type = obj.TYPE_REG
+		p.From.Reg = mips.REGZERO
+		p.To.Type = obj.TYPE_MEM
+		p.To.Name = obj.NAME_AUTO
+		p.To.Reg = mips.REGSP
+		p.To.Offset = n.Xoffset + i
+		p.To.Sym = sym
+	}
+}
+
 func ginsnop() {
 	p := gc.Prog(mips.ANOR)
 	p.From.Type = obj.TYPE_REG

src/cmd/compile/internal/mips64/galign.go

@@ -25,4 +25,5 @@ func Init() {
 	gc.Thearch.SSAMarkMoves = func(s *gc.SSAGenState, b *ssa.Block) {}
 	gc.Thearch.SSAGenValue = ssaGenValue
 	gc.Thearch.SSAGenBlock = ssaGenBlock
+	gc.Thearch.ZeroAuto = zeroAuto
 }

src/cmd/compile/internal/mips64/ggen.go

@@ -95,6 +95,23 @@ func zerorange(p *obj.Prog, frame int64, lo int64, hi int64) *obj.Prog {
 	return p
 }
 
+func zeroAuto(n *gc.Node, pp *obj.Prog) {
+	// Note: this code must not clobber any registers.
+	sym := gc.Linksym(n.Sym)
+	size := n.Type.Size()
+	for i := int64(0); i < size; i += 8 {
+		p := gc.AddAsmAfter(mips.AMOVV, pp)
+		pp = p
+		p.From.Type = obj.TYPE_REG
+		p.From.Reg = mips.REGZERO
+		p.To.Type = obj.TYPE_MEM
+		p.To.Name = obj.NAME_AUTO
+		p.To.Reg = mips.REGSP
+		p.To.Offset = n.Xoffset + i
+		p.To.Sym = sym
+	}
+}
+
 func ginsnop() {
 	p := gc.Prog(mips.ANOR)
 	p.From.Type = obj.TYPE_REG

src/cmd/compile/internal/ppc64/galign.go

@@ -24,6 +24,7 @@ func Init() {
 	gc.Thearch.SSAMarkMoves = ssaMarkMoves
 	gc.Thearch.SSAGenValue = ssaGenValue
 	gc.Thearch.SSAGenBlock = ssaGenBlock
+	gc.Thearch.ZeroAuto = zeroAuto
 
 	initvariants()
 	initproginfo()

src/cmd/compile/internal/ppc64/ggen.go

@@ -90,6 +90,23 @@ func zerorange(p *obj.Prog, frame int64, lo int64, hi int64) *obj.Prog {
 	return p
 }
 
+func zeroAuto(n *gc.Node, pp *obj.Prog) {
+	// Note: this code must not clobber any registers.
+	sym := gc.Linksym(n.Sym)
+	size := n.Type.Size()
+	for i := int64(0); i < size; i += 8 {
+		p := gc.AddAsmAfter(ppc64.AMOVD, pp)
+		pp = p
+		p.From.Type = obj.TYPE_REG
+		p.From.Reg = ppc64.REGZERO
+		p.To.Type = obj.TYPE_MEM
+		p.To.Name = obj.NAME_AUTO
+		p.To.Reg = ppc64.REGSP
+		p.To.Offset = n.Xoffset + i
+		p.To.Sym = sym
+	}
+}
+
 func ginsnop() {
 	p := gc.Prog(ppc64.AOR)
 	p.From.Type = obj.TYPE_REG

src/cmd/compile/internal/s390x/galign.go

@@ -20,4 +20,5 @@ func Init() {
 	gc.Thearch.SSAMarkMoves = ssaMarkMoves
 	gc.Thearch.SSAGenValue = ssaGenValue
 	gc.Thearch.SSAGenBlock = ssaGenBlock
+	gc.Thearch.ZeroAuto = zeroAuto
 }

src/cmd/compile/internal/s390x/ggen.go

@@ -143,6 +143,19 @@ func zerorange(p *obj.Prog, frame int64, lo int64, hi int64) *obj.Prog {
 	return p
 }
 
+func zeroAuto(n *gc.Node, pp *obj.Prog) {
+	// Note: this code must not clobber any registers.
+	p := gc.AddAsmAfter(s390x.ACLEAR, pp)
+	pp = p
+	p.From.Type = obj.TYPE_CONST
+	p.From.Offset = n.Type.Size()
+	p.To.Type = obj.TYPE_MEM
+	p.To.Name = obj.NAME_AUTO
+	p.To.Reg = s390x.REGSP
+	p.To.Offset = n.Xoffset
+	p.To.Sym = gc.Linksym(n.Sym)
+}
+
 func ginsnop() {
 	p := gc.Prog(s390x.AOR)
 	p.From.Type = obj.TYPE_REG

src/cmd/compile/internal/s390x/ssa.go

@@ -424,7 +424,7 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) {
 		p.To.Type = obj.TYPE_MEM
 		p.To.Reg = v.Args[0].Reg()
 		gc.AddAux2(&p.To, v, sc.Off())
-	case ssa.OpCopy, ssa.OpS390XMOVDconvert:
+	case ssa.OpCopy, ssa.OpS390XMOVDconvert, ssa.OpS390XMOVDreg:
 		if v.Type.IsMemory() {
 			return
 		}
@@ -433,6 +433,11 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) {
 		if x != y {
 			opregreg(moveByType(v.Type), y, x)
 		}
+	case ssa.OpS390XMOVDnop:
+		if v.Reg() != v.Args[0].Reg() {
+			v.Fatalf("input[0] and output not in same register %s", v.LongString())
+		}
+		// nothing to do
 	case ssa.OpLoadReg:
 		if v.Type.IsFlags() {
 			v.Fatalf("load flags not implemented: %v", v.LongString())

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

@@ -294,6 +294,39 @@ func checkFunc(f *Func) {
 			}
 		}
 	}
+
+	// Check that if a tuple has a memory type, it is second.
+	for _, b := range f.Blocks {
+		for _, v := range b.Values {
+			if v.Type.IsTuple() && v.Type.FieldType(0).IsMemory() {
+				f.Fatalf("memory is first in a tuple: %s\n", v.LongString())
+			}
+		}
+	}
+
+	// Check that only one memory is live at any point.
+	// TODO: make this check examine interblock.
+	if f.scheduled {
+		for _, b := range f.Blocks {
+			var mem *Value // the live memory
+			for _, v := range b.Values {
+				if v.Op != OpPhi {
+					for _, a := range v.Args {
+						if a.Type.IsMemory() || a.Type.IsTuple() && a.Type.FieldType(1).IsMemory() {
+							if mem == nil {
+								mem = a
+							} else if mem != a {
+								f.Fatalf("two live mems @ %s: %s and %s", v, mem, a)
+							}
+						}
+					}
+				}
+				if v.Type.IsMemory() || v.Type.IsTuple() && v.Type.FieldType(1).IsMemory() {
+					mem = v
+				}
+			}
+		}
+	}
 }
 
 // domCheck reports whether x dominates y (including x==y).

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

@@ -5,6 +5,7 @@
 package ssa
 
 import (
+	"cmd/internal/obj"
 	"fmt"
 	"log"
 	"os"
@@ -349,6 +350,8 @@ var passes = [...]pass{
 	{name: "writebarrier", fn: writebarrier, required: true}, // expand write barrier ops
 	{name: "fuse", fn: fuse},
 	{name: "dse", fn: dse},
+	{name: "insert resched checks", fn: insertLoopReschedChecks,
+		disabled: obj.Preemptibleloops_enabled == 0}, // insert resched checks in loops.
 	{name: "tighten", fn: tighten}, // move values closer to their uses
 	{name: "lower", fn: lower, required: true},
 	{name: "lowered cse", fn: cse},
@@ -378,7 +381,13 @@ type constraint struct {
 }
 
 var passOrder = [...]constraint{
-	// prove reliese on common-subexpression elimination for maximum benefits.
+	// "insert resched checks" uses mem, better to clean out stores first.
+	{"dse", "insert resched checks"},
+	// insert resched checks adds new blocks containing generic instructions
+	{"insert resched checks", "lower"},
+	{"insert resched checks", "tighten"},
+
+	// prove relies on common-subexpression elimination for maximum benefits.
 	{"generic cse", "prove"},
 	// deadcode after prove to eliminate all new dead blocks.
 	{"prove", "generic deadcode"},

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

@@ -313,9 +313,13 @@ func cmpVal(v, w *Value, auxIDs auxmap) Cmp {
 		// that generate memory.
 		return lt2Cmp(v.ID < w.ID)
 	}
-
-	if tc := v.Type.Compare(w.Type); tc != CMPeq {
-		return tc
+	// OpSelect is a pseudo-op. We need to be more agressive
+	// regarding CSE to keep multiple OpSelect's of the same
+	// argument from existing.
+	if v.Op != OpSelect0 && v.Op != OpSelect1 {
+		if tc := v.Type.Compare(w.Type); tc != CMPeq {
+			return tc
+		}
 	}
 
 	if v.Aux != w.Aux {

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

@@ -64,7 +64,7 @@ func liveValues(f *Func, reachable []bool) []bool {
 			q = append(q, v)
 		}
 		for _, v := range b.Values {
-			if opcodeTable[v.Op].call && !live[v.ID] {
+			if (opcodeTable[v.Op].call || opcodeTable[v.Op].hasSideEffects) && !live[v.ID] {
 				live[v.ID] = true
 				q = append(q, v)
 			}

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

@@ -6,6 +6,7 @@ package ssa
 
 import (
 	"cmd/internal/obj"
+	"cmd/internal/obj/s390x"
 	"cmd/internal/obj/x86"
 	"testing"
 )
@@ -21,6 +22,10 @@ func testConfig(t testing.TB) *Config {
 	return NewConfig("amd64", DummyFrontend{t}, testCtxt, true)
 }
 
+func testConfigS390X(t testing.TB) *Config {
+	return NewConfig("s390x", DummyFrontend{t}, obj.Linknew(&s390x.Links390x), true)
+}
+
 // DummyFrontend is a test-only frontend.
 // It assumes 64 bit integers and pointers.
 type DummyFrontend struct {
@@ -30,8 +35,20 @@ type DummyFrontend struct {
 func (DummyFrontend) StringData(s string) interface{} {
 	return nil
 }
-func (DummyFrontend) Auto(t Type) GCNode {
-	return nil
+
+type dummyGCNode struct {
+	typ  Type
+	name string
+}
+
+func (d *dummyGCNode) Typ() Type {
+	return d.typ
+}
+func (d *dummyGCNode) String() string {
+	return d.name
+}
+func (d DummyFrontend) Auto(t Type) GCNode {
+	return &dummyGCNode{typ: t, name: "dummy"}
 }
 func (d DummyFrontend) SplitString(s LocalSlot) (LocalSlot, LocalSlot) {
 	return LocalSlot{s.N, d.TypeBytePtr(), s.Off}, LocalSlot{s.N, d.TypeInt(), s.Off + 8}

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

@@ -24,6 +24,7 @@ type Func struct {
 	vid        idAlloc     // value ID allocator
 
 	scheduled bool // Values in Blocks are in final order
+	NoSplit   bool // true if function is marked as nosplit.  Used by schedule check pass.
 
 	// when register allocation is done, maps value ids to locations
 	RegAlloc []Location

src/cmd/compile/internal/ssa/gen/AMD64Ops.go

@@ -522,15 +522,15 @@ func init() {
 		// store arg0 to arg1+auxint+aux, arg2=mem.
 		// These ops return a tuple of <old contents of *(arg1+auxint+aux), memory>.
 		// Note: arg0 and arg1 are backwards compared to MOVLstore (to facilitate resultInArg0)!
-		{name: "XCHGL", argLength: 3, reg: gpstorexchg, asm: "XCHGL", aux: "SymOff", resultInArg0: true, faultOnNilArg1: true},
-		{name: "XCHGQ", argLength: 3, reg: gpstorexchg, asm: "XCHGQ", aux: "SymOff", resultInArg0: true, faultOnNilArg1: true},
+		{name: "XCHGL", argLength: 3, reg: gpstorexchg, asm: "XCHGL", aux: "SymOff", resultInArg0: true, faultOnNilArg1: true, hasSideEffects: true},
+		{name: "XCHGQ", argLength: 3, reg: gpstorexchg, asm: "XCHGQ", aux: "SymOff", resultInArg0: true, faultOnNilArg1: true, hasSideEffects: true},
 
 		// Atomic adds.
 		// *(arg1+auxint+aux) += arg0.  arg2=mem.
 		// Returns a tuple of <old contents of *(arg1+auxint+aux), memory>.
 		// Note: arg0 and arg1 are backwards compared to MOVLstore (to facilitate resultInArg0)!
-		{name: "XADDLlock", argLength: 3, reg: gpstorexchg, asm: "XADDL", typ: "(UInt32,Mem)", aux: "SymOff", resultInArg0: true, clobberFlags: true, faultOnNilArg1: true},
-		{name: "XADDQlock", argLength: 3, reg: gpstorexchg, asm: "XADDQ", typ: "(UInt64,Mem)", aux: "SymOff", resultInArg0: true, clobberFlags: true, faultOnNilArg1: true},
+		{name: "XADDLlock", argLength: 3, reg: gpstorexchg, asm: "XADDL", typ: "(UInt32,Mem)", aux: "SymOff", resultInArg0: true, clobberFlags: true, faultOnNilArg1: true, hasSideEffects: true},
+		{name: "XADDQlock", argLength: 3, reg: gpstorexchg, asm: "XADDQ", typ: "(UInt64,Mem)", aux: "SymOff", resultInArg0: true, clobberFlags: true, faultOnNilArg1: true, hasSideEffects: true},
 		{name: "AddTupleFirst32", argLength: 2}, // arg0=tuple <x,y>.  Returns <x+arg1,y>.
 		{name: "AddTupleFirst64", argLength: 2}, // arg0=tuple <x,y>.  Returns <x+arg1,y>.
 
@@ -553,12 +553,12 @@ func init() {
 		//    JEQ ...
 		// but we can't do that because memory-using ops can't generate flags yet
 		// (flagalloc wants to move flag-generating instructions around).
-		{name: "CMPXCHGLlock", argLength: 4, reg: cmpxchg, asm: "CMPXCHGL", aux: "SymOff", clobberFlags: true, faultOnNilArg0: true},
-		{name: "CMPXCHGQlock", argLength: 4, reg: cmpxchg, asm: "CMPXCHGQ", aux: "SymOff", clobberFlags: true, faultOnNilArg0: true},
+		{name: "CMPXCHGLlock", argLength: 4, reg: cmpxchg, asm: "CMPXCHGL", aux: "SymOff", clobberFlags: true, faultOnNilArg0: true, hasSideEffects: true},
+		{name: "CMPXCHGQlock", argLength: 4, reg: cmpxchg, asm: "CMPXCHGQ", aux: "SymOff", clobberFlags: true, faultOnNilArg0: true, hasSideEffects: true},
 
 		// Atomic memory updates.
-		{name: "ANDBlock", argLength: 3, reg: gpstore, asm: "ANDB", aux: "SymOff", clobberFlags: true, faultOnNilArg0: true}, // *(arg0+auxint+aux) &= arg1
-		{name: "ORBlock", argLength: 3, reg: gpstore, asm: "ORB", aux: "SymOff", clobberFlags: true, faultOnNilArg0: true},   // *(arg0+auxint+aux) |= arg1
+		{name: "ANDBlock", argLength: 3, reg: gpstore, asm: "ANDB", aux: "SymOff", clobberFlags: true, faultOnNilArg0: true, hasSideEffects: true}, // *(arg0+auxint+aux) &= arg1
+		{name: "ORBlock", argLength: 3, reg: gpstore, asm: "ORB", aux: "SymOff", clobberFlags: true, faultOnNilArg0: true, hasSideEffects: true},   // *(arg0+auxint+aux) |= arg1
 	}
 
 	var AMD64blocks = []blockData{

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

@@ -178,8 +178,8 @@
 (Rsh8x64 x (Const64 [c])) && uint64(c) >= 8 -> (SRAconst (SLLconst <config.fe.TypeUInt32()> x [24]) [31])
 
 (Lrot32 x [c]) -> (SRRconst x [32-c&31])
-(Lrot16 <t> x [c]) -> (OR (SLLconst <t> x [c&15]) (SRLconst <t> x [16-c&15]))
-(Lrot8 <t> x [c]) -> (OR (SLLconst <t> x [c&7]) (SRLconst <t> x [8-c&7]))
+(Lrot16 <t> x [c]) -> (OR (SLLconst <t> x [c&15]) (SRLconst <t> (ZeroExt16to32 x) [16-c&15]))
+(Lrot8 <t> x [c]) -> (OR (SLLconst <t> x [c&7]) (SRLconst <t> (ZeroExt8to32 x) [8-c&7]))
 
 // constants
 (Const8 [val]) -> (MOVWconst [val])

src/cmd/compile/internal/ssa/gen/ARM64.rules

@@ -529,110 +529,148 @@
 // can be encoded in the instructions
 // since this rewriting takes place before stack allocation, the offset to SP is unknown,
 // so don't do it for args and locals with unaligned offset
-(MOVBload [off1] {sym} (ADDconst [off2] ptr) mem) -> (MOVBload [off1+off2] {sym} ptr mem)
-(MOVBUload [off1] {sym} (ADDconst [off2] ptr) mem) -> (MOVBUload [off1+off2] {sym} ptr mem)
+(MOVBload [off1] {sym} (ADDconst [off2] ptr) mem) && is32Bit(off1+off2) ->
+	(MOVBload [off1+off2] {sym} ptr mem)
+(MOVBUload [off1] {sym} (ADDconst [off2] ptr) mem) && is32Bit(off1+off2) ->
+	(MOVBUload [off1+off2] {sym} ptr mem)
 (MOVHload [off1] {sym} (ADDconst [off2] ptr) mem)
-	&& (off1+off2)%2==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym) && !isAuto(sym) ->
+	&& is32Bit(off1+off2) && !isArg(sym)
+	&& ((off1+off2)%2==0 || off1+off2<256 && off1+off2>-256 && !isAuto(sym)) ->
 	(MOVHload [off1+off2] {sym} ptr mem)
 (MOVHUload [off1] {sym} (ADDconst [off2] ptr) mem)
-	&& (off1+off2)%2==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym) && !isAuto(sym) ->
+	&& is32Bit(off1+off2) && !isArg(sym)
+	&& ((off1+off2)%2==0 || off1+off2<256 && off1+off2>-256 && !isAuto(sym)) ->
 	(MOVHUload [off1+off2] {sym} ptr mem)
 (MOVWload [off1] {sym} (ADDconst [off2] ptr) mem)
-	&& (off1+off2)%4==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym) && !isAuto(sym) ->
+	&& is32Bit(off1+off2) && !isArg(sym)
+	&& ((off1+off2)%4==0 || off1+off2<256 && off1+off2>-256 && !isAuto(sym)) ->
 	(MOVWload [off1+off2] {sym} ptr mem)
 (MOVWUload [off1] {sym} (ADDconst [off2] ptr) mem)
-	&& (off1+off2)%4==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym) && !isAuto(sym) ->
+	&& is32Bit(off1+off2) && !isArg(sym)
+	&& ((off1+off2)%4==0 || off1+off2<256 && off1+off2>-256 && !isAuto(sym)) ->
 	(MOVWUload [off1+off2] {sym} ptr mem)
 (MOVDload [off1] {sym} (ADDconst [off2] ptr) mem)
-	&& (off1+off2)%8==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym) && !isAuto(sym) ->
+	&& is32Bit(off1+off2) && !isArg(sym)
+	&& ((off1+off2)%8==0 || off1+off2<256 && off1+off2>-256 && !isAuto(sym)) ->
 	(MOVDload [off1+off2] {sym} ptr mem)
 (FMOVSload [off1] {sym} (ADDconst [off2] ptr) mem)
-	&& (off1+off2)%4==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym) && !isAuto(sym) ->
+	&& is32Bit(off1+off2) && !isArg(sym)
+	&& ((off1+off2)%4==0 || off1+off2<256 && off1+off2>-256 && !isAuto(sym)) ->
 	(FMOVSload [off1+off2] {sym} ptr mem)
 (FMOVDload [off1] {sym} (ADDconst [off2] ptr) mem)
-	&& (off1+off2)%8==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym) && !isAuto(sym) ->
+	&& is32Bit(off1+off2) && !isArg(sym)
+	&& ((off1+off2)%8==0 || off1+off2<256 && off1+off2>-256 && !isAuto(sym)) ->
 	(FMOVDload [off1+off2] {sym} ptr mem)
 
-(MOVBstore [off1] {sym} (ADDconst [off2] ptr) val mem) -> (MOVBstore [off1+off2] {sym} ptr val mem)
+(MOVBstore [off1] {sym} (ADDconst [off2] ptr) val mem) && is32Bit(off1+off2) ->
+	(MOVBstore [off1+off2] {sym} ptr val mem)
 (MOVHstore [off1] {sym} (ADDconst [off2] ptr) val mem)
-	&& (off1+off2)%2==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym) && !isAuto(sym) ->
+	&& is32Bit(off1+off2) && !isArg(sym)
+	&& ((off1+off2)%2==0 || off1+off2<256 && off1+off2>-256 && !isAuto(sym)) ->
 	(MOVHstore [off1+off2] {sym} ptr val mem)
 (MOVWstore [off1] {sym} (ADDconst [off2] ptr) val mem)
-	&& (off1+off2)%4==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym) && !isAuto(sym) ->
+	&& is32Bit(off1+off2) && !isArg(sym)
+	&& ((off1+off2)%4==0 || off1+off2<256 && off1+off2>-256 && !isAuto(sym)) ->
 	(MOVWstore [off1+off2] {sym} ptr val mem)
 (MOVDstore [off1] {sym} (ADDconst [off2] ptr) val mem)
-	&& (off1+off2)%8==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym) && !isAuto(sym) ->
+	&& is32Bit(off1+off2) && !isArg(sym)
+	&& ((off1+off2)%8==0 || off1+off2<256 && off1+off2>-256 && !isAuto(sym)) ->
 	(MOVDstore [off1+off2] {sym} ptr val mem)
 (FMOVSstore [off1] {sym} (ADDconst [off2] ptr) val mem)
-	&& (off1+off2)%4==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym) && !isAuto(sym) ->
+	&& is32Bit(off1+off2) && !isArg(sym)
+	&& ((off1+off2)%4==0 || off1+off2<256 && off1+off2>-256 && !isAuto(sym)) ->
 	(FMOVSstore [off1+off2] {sym} ptr val mem)
 (FMOVDstore [off1] {sym} (ADDconst [off2] ptr) val mem)
-	&& (off1+off2)%8==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym) && !isAuto(sym) ->
+	&& is32Bit(off1+off2) && !isArg(sym)
+	&& ((off1+off2)%8==0 || off1+off2<256 && off1+off2>-256 && !isAuto(sym)) ->
 	(FMOVDstore [off1+off2] {sym} ptr val mem)
-(MOVBstorezero [off1] {sym} (ADDconst [off2] ptr) mem) -> (MOVBstorezero [off1+off2] {sym} ptr mem)
+(MOVBstorezero [off1] {sym} (ADDconst [off2] ptr) mem) && is32Bit(off1+off2) ->
+	(MOVBstorezero [off1+off2] {sym} ptr mem)
 (MOVHstorezero [off1] {sym} (ADDconst [off2] ptr) mem)
-	&& (off1+off2)%2==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym) && !isAuto(sym) ->
+	&& is32Bit(off1+off2) && !isArg(sym)
+	&& ((off1+off2)%2==0 || off1+off2<256 && off1+off2>-256 && !isAuto(sym)) ->
 	(MOVHstorezero [off1+off2] {sym} ptr mem)
 (MOVWstorezero [off1] {sym} (ADDconst [off2] ptr) mem)
-	&& (off1+off2)%4==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym) && !isAuto(sym) ->
+	&& is32Bit(off1+off2) && !isArg(sym)
+	&& ((off1+off2)%4==0 || off1+off2<256 && off1+off2>-256 && !isAuto(sym)) ->
 	(MOVWstorezero [off1+off2] {sym} ptr mem)
 (MOVDstorezero [off1] {sym} (ADDconst [off2] ptr) mem)
-	&& (off1+off2)%2==8 || off1+off2<256 && off1+off2>-256 && !isArg(sym) && !isAuto(sym) ->
+	&& is32Bit(off1+off2) && !isArg(sym)
+	&& ((off1+off2)%2==8 || off1+off2<256 && off1+off2>-256 && !isAuto(sym)) ->
 	(MOVDstorezero [off1+off2] {sym} ptr mem)
 
-(MOVBload [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) ->
+(MOVBload [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2)
+	&& is32Bit(off1+off2) ->
 	(MOVBload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
-(MOVBUload [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) ->
+(MOVBUload [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2)
+	&& is32Bit(off1+off2) ->
 	(MOVBUload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
 (MOVHload [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2)
-	&& ((off1+off2)%2==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym1) && !isAuto(sym1)) ->
+	&& is32Bit(off1+off2) && !isArg(mergeSym(sym1,sym2))
+	&& ((off1+off2)%2==0 || off1+off2<256 && off1+off2>-256 && !isAuto(mergeSym(sym1,sym2))) ->
 	(MOVHload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
 (MOVHUload [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2)
-	&& ((off1+off2)%2==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym1) && !isAuto(sym1)) ->
+	&& is32Bit(off1+off2) && !isArg(mergeSym(sym1,sym2))
+	&& ((off1+off2)%2==0 || off1+off2<256 && off1+off2>-256 && !isAuto(mergeSym(sym1,sym2))) ->
 	(MOVHUload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
 (MOVWload [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2)
-	&& ((off1+off2)%4==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym1) && !isAuto(sym1)) ->
+	&& is32Bit(off1+off2) && !isArg(mergeSym(sym1,sym2))
+	&& ((off1+off2)%4==0 || off1+off2<256 && off1+off2>-256 && !isAuto(mergeSym(sym1,sym2))) ->
 	(MOVWload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
 (MOVWUload [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2)
-	&& ((off1+off2)%4==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym1) && !isAuto(sym1)) ->
+	&& is32Bit(off1+off2) && !isArg(mergeSym(sym1,sym2))
+	&& ((off1+off2)%4==0 || off1+off2<256 && off1+off2>-256 && !isAuto(mergeSym(sym1,sym2))) ->
 	(MOVWUload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
 (MOVDload [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2)
-	&& ((off1+off2)%8==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym1) && !isAuto(sym1)) ->
+	&& is32Bit(off1+off2) && !isArg(mergeSym(sym1,sym2))
+	&& ((off1+off2)%8==0 || off1+off2<256 && off1+off2>-256 && !isAuto(mergeSym(sym1,sym2))) ->
 	(MOVDload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
 (FMOVSload [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2)
-	&& ((off1+off2)%4==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym1) && !isAuto(sym1)) ->
+	&& is32Bit(off1+off2) && !isArg(mergeSym(sym1,sym2))
+	&& ((off1+off2)%4==0 || off1+off2<256 && off1+off2>-256 && !isAuto(mergeSym(sym1,sym2))) ->
 	(FMOVSload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
 (FMOVDload [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2)
-	&& ((off1+off2)%8==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym1) && !isAuto(sym1)) ->
+	&& is32Bit(off1+off2) && !isArg(mergeSym(sym1,sym2))
+	&& ((off1+off2)%8==0 || off1+off2<256 && off1+off2>-256 && !isAuto(mergeSym(sym1,sym2))) ->
 	(FMOVDload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
 
-(MOVBstore [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2) ->
+(MOVBstore [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2)
+	&& is32Bit(off1+off2) ->
 	(MOVBstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem)
 (MOVHstore [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2)
-	&& ((off1+off2)%2==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym1) && !isAuto(sym1)) ->
+	&& is32Bit(off1+off2) && !isArg(mergeSym(sym1,sym2))
+	&& ((off1+off2)%2==0 || off1+off2<256 && off1+off2>-256 && !isAuto(mergeSym(sym1,sym2))) ->
 	(MOVHstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem)
 (MOVWstore [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2)
-	&& ((off1+off2)%4==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym1) && !isAuto(sym1)) ->
+	&& is32Bit(off1+off2) && !isArg(mergeSym(sym1,sym2))
+	&& ((off1+off2)%4==0 || off1+off2<256 && off1+off2>-256 && !isAuto(mergeSym(sym1,sym2))) ->
 	(MOVWstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem)
 (MOVDstore [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2)
-	&& ((off1+off2)%8==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym1) && !isAuto(sym1)) ->
+	&& is32Bit(off1+off2) && !isArg(mergeSym(sym1,sym2))
+	&& ((off1+off2)%8==0 || off1+off2<256 && off1+off2>-256 && !isAuto(mergeSym(sym1,sym2))) ->
 	(MOVDstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem)
 (FMOVSstore [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2)
-	&& ((off1+off2)%4==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym1) && !isAuto(sym1)) ->
+	&& is32Bit(off1+off2) && !isArg(mergeSym(sym1,sym2))
+	&& ((off1+off2)%4==0 || off1+off2<256 && off1+off2>-256 && !isAuto(mergeSym(sym1,sym2))) ->
 	(FMOVSstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem)
 (FMOVDstore [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) val mem) && canMergeSym(sym1,sym2)
-	&& ((off1+off2)%8==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym1) && !isAuto(sym1)) ->
+	&& is32Bit(off1+off2) && !isArg(mergeSym(sym1,sym2))
+	&& ((off1+off2)%8==0 || off1+off2<256 && off1+off2>-256 && !isAuto(mergeSym(sym1,sym2))) ->
 	(FMOVDstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem)
-(MOVBstorezero [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2) ->
+(MOVBstorezero [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2)
+	&& is32Bit(off1+off2) ->
 	(MOVBstorezero [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
 (MOVHstorezero [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2)
-	&& ((off1+off2)%2==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym1) && !isAuto(sym1)) ->
+	&& is32Bit(off1+off2) && !isArg(mergeSym(sym1,sym2))
+	&& ((off1+off2)%2==0 || off1+off2<256 && off1+off2>-256 && !isAuto(mergeSym(sym1,sym2))) ->
 	(MOVHstorezero [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
 (MOVWstorezero [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2)
-	&& ((off1+off2)%4==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym1) && !isAuto(sym1)) ->
+	&& is32Bit(off1+off2) && !isArg(mergeSym(sym1,sym2))
+	&& ((off1+off2)%4==0 || off1+off2<256 && off1+off2>-256 && !isAuto(mergeSym(sym1,sym2))) ->
 	(MOVWstorezero [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
 (MOVDstorezero [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem) && canMergeSym(sym1,sym2)
-	&& ((off1+off2)%8==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym1) && !isAuto(sym1)) ->
+	&& is32Bit(off1+off2) && !isArg(mergeSym(sym1,sym2))
+	&& ((off1+off2)%8==0 || off1+off2<256 && off1+off2>-256 && !isAuto(mergeSym(sym1,sym2))) ->
 	(MOVDstorezero [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
 
 // store zero

src/cmd/compile/internal/ssa/gen/ARM64Ops.go

@@ -456,16 +456,16 @@ func init() {
 
 		// atomic stores.
 		// store arg1 to arg0. arg2=mem. returns memory. auxint must be zero.
-		{name: "STLR", argLength: 3, reg: gpstore, asm: "STLR", faultOnNilArg0: true},
-		{name: "STLRW", argLength: 3, reg: gpstore, asm: "STLRW", faultOnNilArg0: true},
+		{name: "STLR", argLength: 3, reg: gpstore, asm: "STLR", faultOnNilArg0: true, hasSideEffects: true},
+		{name: "STLRW", argLength: 3, reg: gpstore, asm: "STLRW", faultOnNilArg0: true, hasSideEffects: true},
 
 		// atomic exchange.
 		// store arg1 to arg0. arg2=mem. returns <old content of *arg0, memory>. auxint must be zero.
 		// LDAXR	(Rarg0), Rout
 		// STLXR	Rarg1, (Rarg0), Rtmp
 		// CBNZ		Rtmp, -2(PC)
-		{name: "LoweredAtomicExchange64", argLength: 3, reg: gpxchg, resultNotInArgs: true, faultOnNilArg0: true},
-		{name: "LoweredAtomicExchange32", argLength: 3, reg: gpxchg, resultNotInArgs: true, faultOnNilArg0: true},
+		{name: "LoweredAtomicExchange64", argLength: 3, reg: gpxchg, resultNotInArgs: true, faultOnNilArg0: true, hasSideEffects: true},
+		{name: "LoweredAtomicExchange32", argLength: 3, reg: gpxchg, resultNotInArgs: true, faultOnNilArg0: true, hasSideEffects: true},
 
 		// atomic add.
 		// *arg0 += arg1. arg2=mem. returns <new content of *arg0, memory>. auxint must be zero.
@@ -473,8 +473,8 @@ func init() {
 		// ADD		Rarg1, Rout
 		// STLXR	Rout, (Rarg0), Rtmp
 		// CBNZ		Rtmp, -3(PC)
-		{name: "LoweredAtomicAdd64", argLength: 3, reg: gpxchg, resultNotInArgs: true, faultOnNilArg0: true},
-		{name: "LoweredAtomicAdd32", argLength: 3, reg: gpxchg, resultNotInArgs: true, faultOnNilArg0: true},
+		{name: "LoweredAtomicAdd64", argLength: 3, reg: gpxchg, resultNotInArgs: true, faultOnNilArg0: true, hasSideEffects: true},
+		{name: "LoweredAtomicAdd32", argLength: 3, reg: gpxchg, resultNotInArgs: true, faultOnNilArg0: true, hasSideEffects: true},
 
 		// atomic compare and swap.
 		// arg0 = pointer, arg1 = old value, arg2 = new value, arg3 = memory. auxint must be zero.
@@ -490,8 +490,8 @@ func init() {
 		// STLXR	Rarg2, (Rarg0), Rtmp
 		// CBNZ		Rtmp, -4(PC)
 		// CSET		EQ, Rout
-		{name: "LoweredAtomicCas64", argLength: 4, reg: gpcas, resultNotInArgs: true, clobberFlags: true, faultOnNilArg0: true},
-		{name: "LoweredAtomicCas32", argLength: 4, reg: gpcas, resultNotInArgs: true, clobberFlags: true, faultOnNilArg0: true},
+		{name: "LoweredAtomicCas64", argLength: 4, reg: gpcas, resultNotInArgs: true, clobberFlags: true, faultOnNilArg0: true, hasSideEffects: true},
+		{name: "LoweredAtomicCas32", argLength: 4, reg: gpcas, resultNotInArgs: true, clobberFlags: true, faultOnNilArg0: true, hasSideEffects: true},
 
 		// atomic and/or.
 		// *arg0 &= (|=) arg1. arg2=mem. returns memory. auxint must be zero.
@@ -499,8 +499,8 @@ func init() {
 		// AND/OR	Rarg1, Rtmp
 		// STLXRB	Rtmp, (Rarg0), Rtmp
 		// CBNZ		Rtmp, -3(PC)
-		{name: "LoweredAtomicAnd8", argLength: 3, reg: gpstore, asm: "AND", faultOnNilArg0: true},
-		{name: "LoweredAtomicOr8", argLength: 3, reg: gpstore, asm: "ORR", faultOnNilArg0: true},
+		{name: "LoweredAtomicAnd8", argLength: 3, reg: gpstore, asm: "AND", faultOnNilArg0: true, hasSideEffects: true},
+		{name: "LoweredAtomicOr8", argLength: 3, reg: gpstore, asm: "ORR", faultOnNilArg0: true, hasSideEffects: true},
 	}
 
 	blocks := []blockData{

src/cmd/compile/internal/ssa/gen/ARMOps.go

@@ -119,8 +119,8 @@ func init() {
 		fp01      = regInfo{inputs: nil, outputs: []regMask{fp}}
 		fp11      = regInfo{inputs: []regMask{fp}, outputs: []regMask{fp}}
 		fp1flags  = regInfo{inputs: []regMask{fp}}
-		fpgp      = regInfo{inputs: []regMask{fp}, outputs: []regMask{gp}}
-		gpfp      = regInfo{inputs: []regMask{gp}, outputs: []regMask{fp}}
+		fpgp      = regInfo{inputs: []regMask{fp}, outputs: []regMask{gp}, clobbers: buildReg("F15")} // int-float conversion uses F15 as tmp
+		gpfp      = regInfo{inputs: []regMask{gp}, outputs: []regMask{fp}, clobbers: buildReg("F15")}
 		fp21      = regInfo{inputs: []regMask{fp, fp}, outputs: []regMask{fp}}
 		fp2flags  = regInfo{inputs: []regMask{fp, fp}}
 		fpload    = regInfo{inputs: []regMask{gpspsbg}, outputs: []regMask{fp}}

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

@@ -404,8 +404,7 @@
 				(ANDconst  <config.fe.TypeUInt32()> [3] ptr)))
 		(NORconst [0] <config.fe.TypeUInt32()> (SLL <config.fe.TypeUInt32()>
 			(MOVWconst [0xff]) (SLLconst <config.fe.TypeUInt32()> [3]
-				(ANDconst <config.fe.TypeUInt32()> [3]
-					(XORconst <config.fe.TypeUInt32()> [3] ptr)))))) mem)
+				(ANDconst <config.fe.TypeUInt32()> [3] ptr))))) mem)
 
 // AtomicOr8(ptr,val) -> LoweredAtomicOr(ptr&^3,uint32(val) << (((ptr^3) & 3) * 8))
 (AtomicOr8 ptr val mem) && config.BigEndian ->

src/cmd/compile/internal/ssa/gen/MIPSOps.go

@@ -267,8 +267,8 @@ func init() {
 		// SYNC
 		// MOVW	Rarg1, (Rarg0)
 		// SYNC
-		{name: "LoweredAtomicStore", argLength: 3, reg: gpstore, faultOnNilArg0: true},
-		{name: "LoweredAtomicStorezero", argLength: 2, reg: gpstore0, faultOnNilArg0: true},
+		{name: "LoweredAtomicStore", argLength: 3, reg: gpstore, faultOnNilArg0: true, hasSideEffects: true},
+		{name: "LoweredAtomicStorezero", argLength: 2, reg: gpstore0, faultOnNilArg0: true, hasSideEffects: true},
 
 		// atomic exchange.
 		// store arg1 to arg0. arg2=mem. returns <old content of *arg0, memory>.
@@ -278,7 +278,7 @@ func init() {
 		// SC	Rtmp, (Rarg0)
 		// BEQ	Rtmp, -3(PC)
 		// SYNC
-		{name: "LoweredAtomicExchange", argLength: 3, reg: gpxchg, resultNotInArgs: true, faultOnNilArg0: true},
+		{name: "LoweredAtomicExchange", argLength: 3, reg: gpxchg, resultNotInArgs: true, faultOnNilArg0: true, hasSideEffects: true},
 
 		// atomic add.
 		// *arg0 += arg1. arg2=mem. returns <new content of *arg0, memory>.
@@ -289,8 +289,8 @@ func init() {
 		// BEQ	Rtmp, -3(PC)
 		// SYNC
 		// ADDU Rarg1, Rout
-		{name: "LoweredAtomicAdd", argLength: 3, reg: gpxchg, resultNotInArgs: true, faultOnNilArg0: true},
-		{name: "LoweredAtomicAddconst", argLength: 2, reg: regInfo{inputs: []regMask{gpspsbg}, outputs: []regMask{gp}}, aux: "Int32", resultNotInArgs: true, faultOnNilArg0: true},
+		{name: "LoweredAtomicAdd", argLength: 3, reg: gpxchg, resultNotInArgs: true, faultOnNilArg0: true, hasSideEffects: true},
+		{name: "LoweredAtomicAddconst", argLength: 2, reg: regInfo{inputs: []regMask{gpspsbg}, outputs: []regMask{gp}}, aux: "Int32", resultNotInArgs: true, faultOnNilArg0: true, hasSideEffects: true},
 
 		// atomic compare and swap.
 		// arg0 = pointer, arg1 = old value, arg2 = new value, arg3 = memory.
@@ -308,7 +308,7 @@ func init() {
 		// SC	Rout, (Rarg0)
 		// BEQ	Rout, -4(PC)
 		// SYNC
-		{name: "LoweredAtomicCas", argLength: 4, reg: gpcas, resultNotInArgs: true, faultOnNilArg0: true},
+		{name: "LoweredAtomicCas", argLength: 4, reg: gpcas, resultNotInArgs: true, faultOnNilArg0: true, hasSideEffects: true},
 
 		// atomic and/or.
 		// *arg0 &= (|=) arg1. arg2=mem. returns memory.
@@ -318,8 +318,8 @@ func init() {
 		// SC	Rtmp, (Rarg0)
 		// BEQ	Rtmp, -3(PC)
 		// SYNC
-		{name: "LoweredAtomicAnd", argLength: 3, reg: gpstore, asm: "AND", faultOnNilArg0: true},
-		{name: "LoweredAtomicOr", argLength: 3, reg: gpstore, asm: "OR", faultOnNilArg0: true},
+		{name: "LoweredAtomicAnd", argLength: 3, reg: gpstore, asm: "AND", faultOnNilArg0: true, hasSideEffects: true},
+		{name: "LoweredAtomicOr", argLength: 3, reg: gpstore, asm: "OR", faultOnNilArg0: true, hasSideEffects: true},
 
 		// large or unaligned zeroing
 		// arg0 = address of memory to zero (in R1, changed as side effect)

src/cmd/compile/internal/ssa/gen/PPC64.rules

@@ -338,9 +338,9 @@
 (Geq32F x y) -> (FGreaterEqual (FCMPU x y))
 (Geq64F x y) -> (FGreaterEqual (FCMPU x y))
 
-(Geq8U x y)  -> (GreaterEqual (CMPU (ZeroExt8to32 x) (ZeroExt8to32 y)))
-(Geq16U x y) -> (GreaterEqual (CMPU (ZeroExt16to32 x) (ZeroExt16to32 y)))
-(Geq32U x y) -> (GreaterEqual (CMPU x y))
+(Geq8U x y)  -> (GreaterEqual (CMPWU (ZeroExt8to32 x) (ZeroExt8to32 y)))
+(Geq16U x y) -> (GreaterEqual (CMPWU (ZeroExt16to32 x) (ZeroExt16to32 y)))
+(Geq32U x y) -> (GreaterEqual (CMPWU x y))
 (Geq64U x y) -> (GreaterEqual (CMPU x y))
 
 // Absorb pseudo-ops into blocks.

src/cmd/compile/internal/ssa/gen/S390X.rules

@@ -312,9 +312,12 @@
 
 // Lowering loads
 (Load <t> ptr mem) && (is64BitInt(t) || isPtr(t)) -> (MOVDload ptr mem)
-(Load <t> ptr mem) && is32BitInt(t) -> (MOVWZload ptr mem)
-(Load <t> ptr mem) && is16BitInt(t) -> (MOVHZload ptr mem)
-(Load <t> ptr mem) && (t.IsBoolean() || is8BitInt(t)) -> (MOVBZload ptr mem)
+(Load <t> ptr mem) && is32BitInt(t) && isSigned(t) -> (MOVWload ptr mem)
+(Load <t> ptr mem) && is32BitInt(t) && !isSigned(t) -> (MOVWZload ptr mem)
+(Load <t> ptr mem) && is16BitInt(t) && isSigned(t) -> (MOVHload ptr mem)
+(Load <t> ptr mem) && is16BitInt(t) && !isSigned(t) -> (MOVHZload ptr mem)
+(Load <t> ptr mem) && is8BitInt(t) && isSigned(t) -> (MOVBload ptr mem)
+(Load <t> ptr mem) && (t.IsBoolean() || (is8BitInt(t) && !isSigned(t))) -> (MOVBZload ptr mem)
 (Load <t> ptr mem) && is32BitFloat(t) -> (FMOVSload ptr mem)
 (Load <t> ptr mem) && is64BitFloat(t) -> (FMOVDload ptr mem)
 
@@ -437,7 +440,7 @@
 (If (MOVDGTnoinv (MOVDconst [0]) (MOVDconst [1]) cmp) yes no) -> (GTF cmp yes no)
 (If (MOVDGEnoinv (MOVDconst [0]) (MOVDconst [1]) cmp) yes no) -> (GEF cmp yes no)
 
-(If cond yes no) -> (NE (CMPWconst [0] (MOVBZreg cond)) yes no)
+(If cond yes no) -> (NE (CMPWconst [0] (MOVBZreg <config.fe.TypeBool()> cond)) yes no)
 
 // ***************************
 // Above: lowering rules
@@ -445,16 +448,28 @@
 // ***************************
 // TODO: Should the optimizations be a separate pass?
 
+// Fold unnecessary type conversions.
+(MOVDreg <t> x) && t.Compare(x.Type) == CMPeq -> x
+(MOVDnop <t> x) && t.Compare(x.Type) == CMPeq -> x
+
+// Propagate constants through type conversions.
+(MOVDreg (MOVDconst [c])) -> (MOVDconst [c])
+(MOVDnop (MOVDconst [c])) -> (MOVDconst [c])
+
+// If a register move has only 1 use, just use the same register without emitting instruction.
+// MOVDnop doesn't emit instruction, only for ensuring the type.
+(MOVDreg x) && x.Uses == 1 -> (MOVDnop x)
+
 // Fold sign extensions into conditional moves of constants.
 // Designed to remove the MOVBZreg inserted by the If lowering.
-(MOVBZreg x:(MOVDLT (MOVDconst [c]) (MOVDconst [d]) _)) && int64(uint8(c)) == c && int64(uint8(d)) == d -> x
-(MOVBZreg x:(MOVDLE (MOVDconst [c]) (MOVDconst [d]) _)) && int64(uint8(c)) == c && int64(uint8(d)) == d -> x
-(MOVBZreg x:(MOVDGT (MOVDconst [c]) (MOVDconst [d]) _)) && int64(uint8(c)) == c && int64(uint8(d)) == d -> x
-(MOVBZreg x:(MOVDGE (MOVDconst [c]) (MOVDconst [d]) _)) && int64(uint8(c)) == c && int64(uint8(d)) == d -> x
-(MOVBZreg x:(MOVDEQ (MOVDconst [c]) (MOVDconst [d]) _)) && int64(uint8(c)) == c && int64(uint8(d)) == d -> x
-(MOVBZreg x:(MOVDNE (MOVDconst [c]) (MOVDconst [d]) _)) && int64(uint8(c)) == c && int64(uint8(d)) == d -> x
-(MOVBZreg x:(MOVDGTnoinv (MOVDconst [c]) (MOVDconst [d]) _)) && int64(uint8(c)) == c && int64(uint8(d)) == d -> x
-(MOVBZreg x:(MOVDGEnoinv (MOVDconst [c]) (MOVDconst [d]) _)) && int64(uint8(c)) == c && int64(uint8(d)) == d -> x
+(MOVBZreg x:(MOVDLT (MOVDconst [c]) (MOVDconst [d]) _)) && int64(uint8(c)) == c && int64(uint8(d)) == d -> (MOVDreg x)
+(MOVBZreg x:(MOVDLE (MOVDconst [c]) (MOVDconst [d]) _)) && int64(uint8(c)) == c && int64(uint8(d)) == d -> (MOVDreg x)
+(MOVBZreg x:(MOVDGT (MOVDconst [c]) (MOVDconst [d]) _)) && int64(uint8(c)) == c && int64(uint8(d)) == d -> (MOVDreg x)
+(MOVBZreg x:(MOVDGE (MOVDconst [c]) (MOVDconst [d]) _)) && int64(uint8(c)) == c && int64(uint8(d)) == d -> (MOVDreg x)
+(MOVBZreg x:(MOVDEQ (MOVDconst [c]) (MOVDconst [d]) _)) && int64(uint8(c)) == c && int64(uint8(d)) == d -> (MOVDreg x)
+(MOVBZreg x:(MOVDNE (MOVDconst [c]) (MOVDconst [d]) _)) && int64(uint8(c)) == c && int64(uint8(d)) == d -> (MOVDreg x)
+(MOVBZreg x:(MOVDGTnoinv (MOVDconst [c]) (MOVDconst [d]) _)) && int64(uint8(c)) == c && int64(uint8(d)) == d -> (MOVDreg x)
+(MOVBZreg x:(MOVDGEnoinv (MOVDconst [c]) (MOVDconst [d]) _)) && int64(uint8(c)) == c && int64(uint8(d)) == d -> (MOVDreg x)
 
 // Fold boolean tests into blocks.
 (NE (CMPWconst [0] (MOVDLT (MOVDconst [0]) (MOVDconst [1]) cmp)) yes no) -> (LT cmp yes no)
@@ -522,8 +537,8 @@
 (CMP (MOVDconst [c]) x) && is32Bit(c) -> (InvertFlags (CMPconst x [c]))
 (CMPW x (MOVDconst [c])) -> (CMPWconst x [c])
 (CMPW (MOVDconst [c]) x) -> (InvertFlags (CMPWconst x [c]))
-(CMPU x (MOVDconst [c])) && is32Bit(c) -> (CMPUconst x [int64(uint32(c))])
-(CMPU (MOVDconst [c]) x) && is32Bit(c) -> (InvertFlags (CMPUconst x [int64(uint32(c))]))
+(CMPU x (MOVDconst [c])) && isU32Bit(c) -> (CMPUconst x [int64(uint32(c))])
+(CMPU (MOVDconst [c]) x) && isU32Bit(c) -> (InvertFlags (CMPUconst x [int64(uint32(c))]))
 (CMPWU x (MOVDconst [c])) -> (CMPWUconst x [int64(uint32(c))])
 (CMPWU (MOVDconst [c]) x) -> (InvertFlags (CMPWUconst x [int64(uint32(c))]))
 
@@ -572,46 +587,46 @@
 (MOVDNE x y (InvertFlags cmp)) -> (MOVDNE x y cmp)
 
 // don't extend after proper load
-(MOVBreg x:(MOVBload _ _)) -> x
-(MOVBZreg x:(MOVBZload _ _)) -> x
-(MOVHreg x:(MOVBload _ _)) -> x
-(MOVHreg x:(MOVBZload _ _)) -> x
-(MOVHreg x:(MOVHload _ _)) -> x
-(MOVHZreg x:(MOVBZload _ _)) -> x
-(MOVHZreg x:(MOVHZload _ _)) -> x
-(MOVWreg x:(MOVBload _ _)) -> x
-(MOVWreg x:(MOVBZload _ _)) -> x
-(MOVWreg x:(MOVHload _ _)) -> x
-(MOVWreg x:(MOVHZload _ _)) -> x
-(MOVWreg x:(MOVWload _ _)) -> x
-(MOVWZreg x:(MOVBZload _ _)) -> x
-(MOVWZreg x:(MOVHZload _ _)) -> x
-(MOVWZreg x:(MOVWZload _ _)) -> x
+(MOVBreg x:(MOVBload _ _)) -> (MOVDreg x)
+(MOVBZreg x:(MOVBZload _ _)) -> (MOVDreg x)
+(MOVHreg x:(MOVBload _ _)) -> (MOVDreg x)
+(MOVHreg x:(MOVBZload _ _)) -> (MOVDreg x)
+(MOVHreg x:(MOVHload _ _)) -> (MOVDreg x)
+(MOVHZreg x:(MOVBZload _ _)) -> (MOVDreg x)
+(MOVHZreg x:(MOVHZload _ _)) -> (MOVDreg x)
+(MOVWreg x:(MOVBload _ _)) -> (MOVDreg x)
+(MOVWreg x:(MOVBZload _ _)) -> (MOVDreg x)
+(MOVWreg x:(MOVHload _ _)) -> (MOVDreg x)
+(MOVWreg x:(MOVHZload _ _)) -> (MOVDreg x)
+(MOVWreg x:(MOVWload _ _)) -> (MOVDreg x)
+(MOVWZreg x:(MOVBZload _ _)) -> (MOVDreg x)
+(MOVWZreg x:(MOVHZload _ _)) -> (MOVDreg x)
+(MOVWZreg x:(MOVWZload _ _)) -> (MOVDreg x)
 
 // don't extend if argument is already extended
-(MOVBreg x:(Arg <t>)) && is8BitInt(t) && isSigned(t) -> x
-(MOVBZreg x:(Arg <t>)) && is8BitInt(t) && !isSigned(t) -> x
-(MOVHreg x:(Arg <t>)) && (is8BitInt(t) || is16BitInt(t)) && isSigned(t) -> x
-(MOVHZreg x:(Arg <t>)) && (is8BitInt(t) || is16BitInt(t)) && !isSigned(t) -> x
-(MOVWreg x:(Arg <t>)) && (is8BitInt(t) || is16BitInt(t) || is32BitInt(t)) && isSigned(t) -> x
-(MOVWZreg x:(Arg <t>)) && (is8BitInt(t) || is16BitInt(t) || is32BitInt(t)) && !isSigned(t) -> x
+(MOVBreg x:(Arg <t>)) && is8BitInt(t) && isSigned(t) -> (MOVDreg x)
+(MOVBZreg x:(Arg <t>)) && is8BitInt(t) && !isSigned(t) -> (MOVDreg x)
+(MOVHreg x:(Arg <t>)) && (is8BitInt(t) || is16BitInt(t)) && isSigned(t) -> (MOVDreg x)
+(MOVHZreg x:(Arg <t>)) && (is8BitInt(t) || is16BitInt(t)) && !isSigned(t) -> (MOVDreg x)
+(MOVWreg x:(Arg <t>)) && (is8BitInt(t) || is16BitInt(t) || is32BitInt(t)) && isSigned(t) -> (MOVDreg x)
+(MOVWZreg x:(Arg <t>)) && (is8BitInt(t) || is16BitInt(t) || is32BitInt(t)) && !isSigned(t) -> (MOVDreg x)
 
 // fold double extensions
-(MOVBreg x:(MOVBreg _)) -> x
-(MOVBZreg x:(MOVBZreg _)) -> x
-(MOVHreg x:(MOVBreg _)) -> x
-(MOVHreg x:(MOVBZreg _)) -> x
-(MOVHreg x:(MOVHreg _)) -> x
-(MOVHZreg x:(MOVBZreg _)) -> x
-(MOVHZreg x:(MOVHZreg _)) -> x
-(MOVWreg x:(MOVBreg _)) -> x
-(MOVWreg x:(MOVBZreg _)) -> x
-(MOVWreg x:(MOVHreg _)) -> x
-(MOVWreg x:(MOVHreg _)) -> x
-(MOVWreg x:(MOVWreg _)) -> x
-(MOVWZreg x:(MOVBZreg _)) -> x
-(MOVWZreg x:(MOVHZreg _)) -> x
-(MOVWZreg x:(MOVWZreg _)) -> x
+(MOVBreg x:(MOVBreg _)) -> (MOVDreg x)
+(MOVBZreg x:(MOVBZreg _)) -> (MOVDreg x)
+(MOVHreg x:(MOVBreg _)) -> (MOVDreg x)
+(MOVHreg x:(MOVBZreg _)) -> (MOVDreg x)
+(MOVHreg x:(MOVHreg _)) -> (MOVDreg x)
+(MOVHZreg x:(MOVBZreg _)) -> (MOVDreg x)
+(MOVHZreg x:(MOVHZreg _)) -> (MOVDreg x)
+(MOVWreg x:(MOVBreg _)) -> (MOVDreg x)
+(MOVWreg x:(MOVBZreg _)) -> (MOVDreg x)
+(MOVWreg x:(MOVHreg _)) -> (MOVDreg x)
+(MOVWreg x:(MOVHreg _)) -> (MOVDreg x)
+(MOVWreg x:(MOVWreg _)) -> (MOVDreg x)
+(MOVWZreg x:(MOVBZreg _)) -> (MOVDreg x)
+(MOVWZreg x:(MOVHZreg _)) -> (MOVDreg x)
+(MOVWZreg x:(MOVWZreg _)) -> (MOVDreg x)
 
 // fold extensions into constants
 (MOVBreg (MOVDconst [c])) -> (MOVDconst [int64(int8(c))])
@@ -641,10 +656,10 @@
 (MOVWZreg x:(MOVWZloadidx [off] {sym} ptr idx mem)) && x.Uses == 1 && clobber(x) -> @x.Block (MOVWZloadidx <v.Type> [off] {sym} ptr idx mem)
 
 // replace load from same location as preceding store with copy
-(MOVBZload [off] {sym} ptr (MOVBstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> x
-(MOVHZload [off] {sym} ptr (MOVHstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> x
-(MOVWZload [off] {sym} ptr (MOVWstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> x
-(MOVDload [off] {sym} ptr (MOVDstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> x
+(MOVBZload [off] {sym} ptr (MOVBstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVDreg x)
+(MOVHZload [off] {sym} ptr (MOVHstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVDreg x)
+(MOVWZload [off] {sym} ptr (MOVWstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVDreg x)
+(MOVDload [off] {sym} ptr (MOVDstore [off2] {sym2} ptr2 x _)) && sym == sym2 && off == off2 && isSamePtr(ptr, ptr2) -> (MOVDreg x)
 
 // Don't extend before storing
 (MOVWstore [off] {sym} ptr (MOVWreg x) mem) -> (MOVWstore [off] {sym} ptr x mem)
@@ -885,9 +900,9 @@
 (MOVDEQ y _ (FlagLT)) -> y
 (MOVDEQ y _ (FlagGT)) -> y
 
-(MOVDNE _ y (FlagEQ)) -> y
-(MOVDNE x _ (FlagLT)) -> x
-(MOVDNE x _ (FlagGT)) -> x
+(MOVDNE y _ (FlagEQ)) -> y
+(MOVDNE _ x (FlagLT)) -> x
+(MOVDNE _ x (FlagGT)) -> x
 
 (MOVDLT y _ (FlagEQ)) -> y
 (MOVDLT _ x (FlagLT)) -> x

src/cmd/compile/internal/ssa/gen/S390XOps.go

@@ -311,6 +311,9 @@ func init() {
 		{name: "MOVHZreg", argLength: 1, reg: gp11sp, asm: "MOVHZ", typ: "UInt64"}, // zero extend arg0 from int16 to int64
 		{name: "MOVWreg", argLength: 1, reg: gp11sp, asm: "MOVW", typ: "Int64"},    // sign extend arg0 from int32 to int64
 		{name: "MOVWZreg", argLength: 1, reg: gp11sp, asm: "MOVWZ", typ: "UInt64"}, // zero extend arg0 from int32 to int64
+		{name: "MOVDreg", argLength: 1, reg: gp11sp, asm: "MOVD"},                  // move from arg0
+
+		{name: "MOVDnop", argLength: 1, reg: gp11, resultInArg0: true}, // nop, return arg0 in same register
 
 		{name: "MOVDconst", reg: gp01, asm: "MOVD", typ: "UInt64", aux: "Int64", rematerializeable: true}, // auxint
 
@@ -426,14 +429,14 @@ func init() {
 
 		// Atomic stores. These are just normal stores.
 		// store arg1 to arg0+auxint+aux. arg2=mem.
-		{name: "MOVWatomicstore", argLength: 3, reg: gpstore, asm: "MOVW", aux: "SymOff", typ: "Mem", clobberFlags: true, faultOnNilArg0: true},
-		{name: "MOVDatomicstore", argLength: 3, reg: gpstore, asm: "MOVD", aux: "SymOff", typ: "Mem", clobberFlags: true, faultOnNilArg0: true},
+		{name: "MOVWatomicstore", argLength: 3, reg: gpstore, asm: "MOVW", aux: "SymOff", typ: "Mem", clobberFlags: true, faultOnNilArg0: true, hasSideEffects: true},
+		{name: "MOVDatomicstore", argLength: 3, reg: gpstore, asm: "MOVD", aux: "SymOff", typ: "Mem", clobberFlags: true, faultOnNilArg0: true, hasSideEffects: true},
 
 		// Atomic adds.
 		// *(arg0+auxint+aux) += arg1.  arg2=mem.
 		// Returns a tuple of <old contents of *(arg0+auxint+aux), memory>.
-		{name: "LAA", argLength: 3, reg: gpstorelaa, asm: "LAA", typ: "(UInt32,Mem)", aux: "SymOff", faultOnNilArg0: true},
-		{name: "LAAG", argLength: 3, reg: gpstorelaa, asm: "LAAG", typ: "(UInt64,Mem)", aux: "SymOff", faultOnNilArg0: true},
+		{name: "LAA", argLength: 3, reg: gpstorelaa, asm: "LAA", typ: "(UInt32,Mem)", aux: "SymOff", faultOnNilArg0: true, hasSideEffects: true},
+		{name: "LAAG", argLength: 3, reg: gpstorelaa, asm: "LAAG", typ: "(UInt64,Mem)", aux: "SymOff", faultOnNilArg0: true, hasSideEffects: true},
 		{name: "AddTupleFirst32", argLength: 2}, // arg0=tuple <x,y>.  Returns <x+arg1,y>.
 		{name: "AddTupleFirst64", argLength: 2}, // arg0=tuple <x,y>.  Returns <x+arg1,y>.
 
@@ -458,13 +461,13 @@ func init() {
 		//    BEQ ...
 		// but we can't do that because memory-using ops can't generate flags yet
 		// (flagalloc wants to move flag-generating instructions around).
-		{name: "LoweredAtomicCas32", argLength: 4, reg: cas, asm: "CS", aux: "SymOff", clobberFlags: true, faultOnNilArg0: true},
-		{name: "LoweredAtomicCas64", argLength: 4, reg: cas, asm: "CSG", aux: "SymOff", clobberFlags: true, faultOnNilArg0: true},
+		{name: "LoweredAtomicCas32", argLength: 4, reg: cas, asm: "CS", aux: "SymOff", clobberFlags: true, faultOnNilArg0: true, hasSideEffects: true},
+		{name: "LoweredAtomicCas64", argLength: 4, reg: cas, asm: "CSG", aux: "SymOff", clobberFlags: true, faultOnNilArg0: true, hasSideEffects: true},
 
 		// Lowered atomic swaps, emulated using compare-and-swap.
 		// store arg1 to arg0+auxint+aux, arg2=mem.
-		{name: "LoweredAtomicExchange32", argLength: 3, reg: exchange, asm: "CS", aux: "SymOff", clobberFlags: true, faultOnNilArg0: true},
-		{name: "LoweredAtomicExchange64", argLength: 3, reg: exchange, asm: "CSG", aux: "SymOff", clobberFlags: true, faultOnNilArg0: true},
+		{name: "LoweredAtomicExchange32", argLength: 3, reg: exchange, asm: "CS", aux: "SymOff", clobberFlags: true, faultOnNilArg0: true, hasSideEffects: true},
+		{name: "LoweredAtomicExchange64", argLength: 3, reg: exchange, asm: "CSG", aux: "SymOff", clobberFlags: true, faultOnNilArg0: true, hasSideEffects: true},
 
 		// find leftmost one
 		{

src/cmd/compile/internal/ssa/gen/genericOps.go

@@ -460,20 +460,20 @@ var genericOps = []opData{
 	// Atomic loads return a new memory so that the loads are properly ordered
 	// with respect to other loads and stores.
 	// TODO: use for sync/atomic at some point.
-	{name: "AtomicLoad32", argLength: 2, typ: "(UInt32,Mem)"},         // Load from arg0.  arg1=memory.  Returns loaded value and new memory.
-	{name: "AtomicLoad64", argLength: 2, typ: "(UInt64,Mem)"},         // Load from arg0.  arg1=memory.  Returns loaded value and new memory.
-	{name: "AtomicLoadPtr", argLength: 2, typ: "(BytePtr,Mem)"},       // Load from arg0.  arg1=memory.  Returns loaded value and new memory.
-	{name: "AtomicStore32", argLength: 3, typ: "Mem"},                 // Store arg1 to *arg0.  arg2=memory.  Returns memory.
-	{name: "AtomicStore64", argLength: 3, typ: "Mem"},                 // Store arg1 to *arg0.  arg2=memory.  Returns memory.
-	{name: "AtomicStorePtrNoWB", argLength: 3, typ: "Mem"},            // Store arg1 to *arg0.  arg2=memory.  Returns memory.
-	{name: "AtomicExchange32", argLength: 3, typ: "(UInt32,Mem)"},     // Store arg1 to *arg0.  arg2=memory.  Returns old contents of *arg0 and new memory.
-	{name: "AtomicExchange64", argLength: 3, typ: "(UInt64,Mem)"},     // Store arg1 to *arg0.  arg2=memory.  Returns old contents of *arg0 and new memory.
-	{name: "AtomicAdd32", argLength: 3, typ: "(UInt32,Mem)"},          // Do *arg0 += arg1.  arg2=memory.  Returns sum and new memory.
-	{name: "AtomicAdd64", argLength: 3, typ: "(UInt64,Mem)"},          // Do *arg0 += arg1.  arg2=memory.  Returns sum and new memory.
-	{name: "AtomicCompareAndSwap32", argLength: 4, typ: "(Bool,Mem)"}, // if *arg0==arg1, then set *arg0=arg2.  Returns true iff store happens and new memory.
-	{name: "AtomicCompareAndSwap64", argLength: 4, typ: "(Bool,Mem)"}, // if *arg0==arg1, then set *arg0=arg2.  Returns true iff store happens and new memory.
-	{name: "AtomicAnd8", argLength: 3, typ: "Mem"},                    // *arg0 &= arg1.  arg2=memory.  Returns memory.
-	{name: "AtomicOr8", argLength: 3, typ: "Mem"},                     // *arg0 |= arg1.  arg2=memory.  Returns memory.
+	{name: "AtomicLoad32", argLength: 2, typ: "(UInt32,Mem)"},                               // Load from arg0.  arg1=memory.  Returns loaded value and new memory.
+	{name: "AtomicLoad64", argLength: 2, typ: "(UInt64,Mem)"},                               // Load from arg0.  arg1=memory.  Returns loaded value and new memory.
+	{name: "AtomicLoadPtr", argLength: 2, typ: "(BytePtr,Mem)"},                             // Load from arg0.  arg1=memory.  Returns loaded value and new memory.
+	{name: "AtomicStore32", argLength: 3, typ: "Mem", hasSideEffects: true},                 // Store arg1 to *arg0.  arg2=memory.  Returns memory.
+	{name: "AtomicStore64", argLength: 3, typ: "Mem", hasSideEffects: true},                 // Store arg1 to *arg0.  arg2=memory.  Returns memory.
+	{name: "AtomicStorePtrNoWB", argLength: 3, typ: "Mem", hasSideEffects: true},            // Store arg1 to *arg0.  arg2=memory.  Returns memory.
+	{name: "AtomicExchange32", argLength: 3, typ: "(UInt32,Mem)", hasSideEffects: true},     // Store arg1 to *arg0.  arg2=memory.  Returns old contents of *arg0 and new memory.
+	{name: "AtomicExchange64", argLength: 3, typ: "(UInt64,Mem)", hasSideEffects: true},     // Store arg1 to *arg0.  arg2=memory.  Returns old contents of *arg0 and new memory.
+	{name: "AtomicAdd32", argLength: 3, typ: "(UInt32,Mem)", hasSideEffects: true},          // Do *arg0 += arg1.  arg2=memory.  Returns sum and new memory.
+	{name: "AtomicAdd64", argLength: 3, typ: "(UInt64,Mem)", hasSideEffects: true},          // Do *arg0 += arg1.  arg2=memory.  Returns sum and new memory.
+	{name: "AtomicCompareAndSwap32", argLength: 4, typ: "(Bool,Mem)", hasSideEffects: true}, // if *arg0==arg1, then set *arg0=arg2.  Returns true iff store happens and new memory.
+	{name: "AtomicCompareAndSwap64", argLength: 4, typ: "(Bool,Mem)", hasSideEffects: true}, // if *arg0==arg1, then set *arg0=arg2.  Returns true iff store happens and new memory.
+	{name: "AtomicAnd8", argLength: 3, typ: "Mem", hasSideEffects: true},                    // *arg0 &= arg1.  arg2=memory.  Returns memory.
+	{name: "AtomicOr8", argLength: 3, typ: "Mem", hasSideEffects: true},                     // *arg0 |= arg1.  arg2=memory.  Returns memory.
 }
 
 //     kind           control    successors       implicit exit

src/cmd/compile/internal/ssa/gen/main.go

@@ -52,6 +52,7 @@ type opData struct {
 	faultOnNilArg0    bool  // this op will fault if arg0 is nil (and aux encodes a small offset)
 	faultOnNilArg1    bool  // this op will fault if arg1 is nil (and aux encodes a small offset)
 	usesScratch       bool  // this op requires scratch memory space
+	hasSideEffects    bool  // for "reasons", not to be eliminated.  E.g., atomic store, #19182.
 }
 
 type blockData struct {
@@ -208,6 +209,9 @@ func genOp() {
 			if v.usesScratch {
 				fmt.Fprintln(w, "usesScratch: true,")
 			}
+			if v.hasSideEffects {
+				fmt.Fprintln(w, "hasSideEffects: true,")
+			}
 			if a.name == "generic" {
 				fmt.Fprintln(w, "generic:true,")
 				fmt.Fprintln(w, "},") // close op

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

@@ -33,13 +33,24 @@ type loop struct {
 
 // outerinner records that outer contains inner
 func (sdom SparseTree) outerinner(outer, inner *loop) {
+	// There could be other outer loops found in some random order,
+	// locate the new outer loop appropriately among them.
 	oldouter := inner.outer
-	if oldouter == nil || sdom.isAncestorEq(oldouter.header, outer.header) {
-		inner.outer = outer
-		outer.isInner = false
-		if inner.containsCall {
-			outer.setContainsCall()
-		}
+	for oldouter != nil && sdom.isAncestor(outer.header, oldouter.header) {
+		inner = oldouter
+		oldouter = inner.outer
+	}
+	if outer == oldouter {
+		return
+	}
+	if oldouter != nil {
+		outer.outer = oldouter
+	}
+
+	inner.outer = outer
+	outer.isInner = false
+	if inner.containsCall {
+		outer.setContainsCall()
 	}
 }
 

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

@@ -0,0 +1,87 @@
+// Copyright 2017 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package ssa
+
+import (
+	"testing"
+)
+
+func TestLoopConditionS390X(t *testing.T) {
+	// Test that a simple loop condition does not generate a conditional
+	// move (issue #19227).
+	//
+	// MOVDLT is generated when Less64 is lowered but should be
+	// optimized into an LT branch.
+	//
+	// For example, compiling the following loop:
+	//
+	//   for i := 0; i < N; i++ {
+	//     sum += 3
+	//   }
+	//
+	// should generate assembly similar to:
+	//   loop:
+	//     CMP    R0, R1
+	//     BGE    done
+	//     ADD    $3, R4
+	//     ADD    $1, R1
+	//     BR     loop
+	//   done:
+	//
+	// rather than:
+	// loop:
+	//     MOVD   $0, R2
+	//     MOVD   $1, R3
+	//     CMP    R0, R1
+	//     MOVDLT R2, R3
+	//     CMPW   R2, $0
+	//     BNE    done
+	//     ADD    $3, R4
+	//     ADD    $1, R1
+	//     BR     loop
+	//   done:
+	//
+	c := testConfigS390X(t)
+	fun := Fun(c, "entry",
+		Bloc("entry",
+			Valu("mem", OpInitMem, TypeMem, 0, nil),
+			Valu("SP", OpSP, TypeUInt64, 0, nil),
+			Valu("Nptr", OpOffPtr, TypeInt64Ptr, 8, nil, "SP"),
+			Valu("ret", OpOffPtr, TypeInt64Ptr, 16, nil, "SP"),
+			Valu("N", OpLoad, TypeInt64, 0, nil, "Nptr", "mem"),
+			Valu("starti", OpConst64, TypeInt64, 0, nil),
+			Valu("startsum", OpConst64, TypeInt64, 0, nil),
+			Goto("b1")),
+		Bloc("b1",
+			Valu("phii", OpPhi, TypeInt64, 0, nil, "starti", "i"),
+			Valu("phisum", OpPhi, TypeInt64, 0, nil, "startsum", "sum"),
+			Valu("cmp1", OpLess64, TypeBool, 0, nil, "phii", "N"),
+			If("cmp1", "b2", "b3")),
+		Bloc("b2",
+			Valu("c1", OpConst64, TypeInt64, 1, nil),
+			Valu("i", OpAdd64, TypeInt64, 0, nil, "phii", "c1"),
+			Valu("c3", OpConst64, TypeInt64, 3, nil),
+			Valu("sum", OpAdd64, TypeInt64, 0, nil, "phisum", "c3"),
+			Goto("b1")),
+		Bloc("b3",
+			Valu("store", OpStore, TypeMem, 8, nil, "ret", "phisum", "mem"),
+			Exit("store")))
+	CheckFunc(fun.f)
+	Compile(fun.f)
+	CheckFunc(fun.f)
+
+	checkOpcodeCounts(t, fun.f, map[Op]int{
+		OpS390XMOVDLT:    0,
+		OpS390XMOVDGT:    0,
+		OpS390XMOVDLE:    0,
+		OpS390XMOVDGE:    0,
+		OpS390XMOVDEQ:    0,
+		OpS390XMOVDNE:    0,
+		OpS390XCMP:       1,
+		OpS390XCMPWconst: 0,
+	})
+
+	fun.f.Free()
+}

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

@@ -0,0 +1,517 @@
+// Copyright 2016 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package ssa
+
+import "fmt"
+
+// an edgeMemCtr records a backedge, together with the memory and
+// counter phi functions at the target of the backedge that must
+// be updated when a rescheduling check replaces the backedge.
+type edgeMemCtr struct {
+	e Edge
+	m *Value // phi for memory at dest of e
+	c *Value // phi for counter at dest of e
+}
+
+// a rewriteTarget is a a value-argindex pair indicating
+// where a rewrite is applied.  Note that this is for values,
+// not for block controls, because block controls are not targets
+// for the rewrites performed in inserting rescheduling checks.
+type rewriteTarget struct {
+	v *Value
+	i int
+}
+
+type rewrite struct {
+	before, after *Value          // before is the expected value before rewrite, after is the new value installed.
+	rewrites      []rewriteTarget // all the targets for this rewrite.
+}
+
+func (r *rewrite) String() string {
+	s := "\n\tbefore=" + r.before.String() + ", after=" + r.after.String()
+	for _, rw := range r.rewrites {
+		s += ", (i=" + fmt.Sprint(rw.i) + ", v=" + rw.v.LongString() + ")"
+	}
+	s += "\n"
+	return s
+}
+
+const initialRescheduleCounterValue = 1021 // Largest 10-bit prime. 97 nSec loop bodies will check every 100 uSec.
+
+// insertLoopReschedChecks inserts rescheduling checks on loop backedges.
+func insertLoopReschedChecks(f *Func) {
+	// TODO: when split information is recorded in export data, insert checks only on backedges that can be reached on a split-call-free path.
+
+	// Loop reschedule checks decrement a per-function counter
+	// shared by all loops, and when the counter becomes non-positive
+	// a call is made to a rescheduling check in the runtime.
+	//
+	// Steps:
+	// 1. locate backedges.
+	// 2. Record memory definitions at block end so that
+	//    the SSA graph for mem can be prperly modified.
+	// 3. Define a counter and record its future uses (at backedges)
+	//    (Same process as 2, applied to a single definition of the counter.
+	//     difference for mem is that there are zero-to-many existing mem
+	//     definitions, versus exactly one for the new counter.)
+	// 4. Ensure that phi functions that will-be-needed for mem and counter
+	//    are present in the graph, initially with trivial inputs.
+	// 5. Record all to-be-modified uses of mem and counter;
+	//    apply modifications (split into two steps to simplify and
+	//    avoided nagging order-dependences).
+	// 6. Rewrite backedges to include counter check, reschedule check,
+	//    and modify destination phi function appropriately with new
+	//    definitions for mem and counter.
+
+	if f.NoSplit { // nosplit functions don't reschedule.
+		return
+	}
+
+	backedges := backedges(f)
+	if len(backedges) == 0 { // no backedges means no rescheduling checks.
+		return
+	}
+
+	lastMems := findLastMems(f)
+
+	idom := f.Idom()
+	sdom := f.sdom()
+
+	if f.pass.debug > 2 {
+		fmt.Printf("before %s = %s\n", f.Name, sdom.treestructure(f.Entry))
+	}
+
+	tofixBackedges := []edgeMemCtr{}
+
+	for _, e := range backedges { // TODO: could filter here by calls in loops, if declared and inferred nosplit are recorded in export data.
+		tofixBackedges = append(tofixBackedges, edgeMemCtr{e, nil, nil})
+	}
+
+	// It's possible that there is no memory state (no global/pointer loads/stores or calls)
+	if lastMems[f.Entry.ID] == nil {
+		lastMems[f.Entry.ID] = f.Entry.NewValue0(f.Entry.Line, OpInitMem, TypeMem)
+	}
+
+	memDefsAtBlockEnds := make([]*Value, f.NumBlocks()) // For each block, the mem def seen at its bottom. Could be from earlier block.
+
+	// Propagate last mem definitions forward through successor blocks.
+	po := f.postorder()
+	for i := len(po) - 1; i >= 0; i-- {
+		b := po[i]
+		mem := lastMems[b.ID]
+		for j := 0; mem == nil; j++ { // if there's no def, then there's no phi, so the visible mem is identical in all predecessors.
+			// loop because there might be backedges that haven't been visited yet.
+			mem = memDefsAtBlockEnds[b.Preds[j].b.ID]
+		}
+		memDefsAtBlockEnds[b.ID] = mem
+	}
+
+	// Set up counter.  There are no phis etc pre-existing for it.
+	counter0 := f.Entry.NewValue0I(f.Entry.Line, OpConst32, f.Config.fe.TypeInt32(), initialRescheduleCounterValue)
+	ctrDefsAtBlockEnds := make([]*Value, f.NumBlocks()) // For each block, def visible at its end, if that def will be used.
+
+	// There's a minor difference between memDefsAtBlockEnds and ctrDefsAtBlockEnds;
+	// because the counter only matter for loops and code that reaches them, it is nil for blocks where the ctr is no
+	// longer live.  This will avoid creation of dead phi functions.  This optimization is ignored for the mem variable
+	// because it is harder and also less likely to be helpful, though dead code elimination ought to clean this out anyhow.
+
+	for _, emc := range tofixBackedges {
+		e := emc.e
+		// set initial uses of counter zero (note available-at-bottom and use are the same thing initially.)
+		// each back-edge will be rewritten to include a reschedule check, and that will use the counter.
+		src := e.b.Preds[e.i].b
+		ctrDefsAtBlockEnds[src.ID] = counter0
+	}
+
+	// Push uses towards root
+	for _, b := range f.postorder() {
+		bd := ctrDefsAtBlockEnds[b.ID]
+		if bd == nil {
+			continue
+		}
+		for _, e := range b.Preds {
+			p := e.b
+			if ctrDefsAtBlockEnds[p.ID] == nil {
+				ctrDefsAtBlockEnds[p.ID] = bd
+			}
+		}
+	}
+
+	// Maps from block to newly-inserted phi function in block.
+	newmemphis := make(map[*Block]rewrite)
+	newctrphis := make(map[*Block]rewrite)
+
+	// Insert phi functions as necessary for future changes to flow graph.
+	for i, emc := range tofixBackedges {
+		e := emc.e
+		h := e.b
+
+		// find the phi function for the memory input at "h", if there is one.
+		var headerMemPhi *Value // look for header mem phi
+
+		for _, v := range h.Values {
+			if v.Op == OpPhi && v.Type.IsMemory() {
+				headerMemPhi = v
+			}
+		}
+
+		if headerMemPhi == nil {
+			// if the header is nil, make a trivial phi from the dominator
+			mem0 := memDefsAtBlockEnds[idom[h.ID].ID]
+			headerMemPhi = newPhiFor(h, mem0)
+			newmemphis[h] = rewrite{before: mem0, after: headerMemPhi}
+			addDFphis(mem0, h, h, f, memDefsAtBlockEnds, newmemphis)
+
+		}
+		tofixBackedges[i].m = headerMemPhi
+
+		var headerCtrPhi *Value
+		rw, ok := newctrphis[h]
+		if !ok {
+			headerCtrPhi = newPhiFor(h, counter0)
+			newctrphis[h] = rewrite{before: counter0, after: headerCtrPhi}
+			addDFphis(counter0, h, h, f, ctrDefsAtBlockEnds, newctrphis)
+		} else {
+			headerCtrPhi = rw.after
+		}
+		tofixBackedges[i].c = headerCtrPhi
+	}
+
+	rewriteNewPhis(f.Entry, f.Entry, f, memDefsAtBlockEnds, newmemphis)
+	rewriteNewPhis(f.Entry, f.Entry, f, ctrDefsAtBlockEnds, newctrphis)
+
+	if f.pass.debug > 0 {
+		for b, r := range newmemphis {
+			fmt.Printf("b=%s, rewrite=%s\n", b, r.String())
+		}
+
+		for b, r := range newctrphis {
+			fmt.Printf("b=%s, rewrite=%s\n", b, r.String())
+		}
+	}
+
+	// Apply collected rewrites.
+	for _, r := range newmemphis {
+		for _, rw := range r.rewrites {
+			rw.v.SetArg(rw.i, r.after)
+		}
+	}
+
+	for _, r := range newctrphis {
+		for _, rw := range r.rewrites {
+			rw.v.SetArg(rw.i, r.after)
+		}
+	}
+
+	zero := f.Entry.NewValue0I(f.Entry.Line, OpConst32, f.Config.fe.TypeInt32(), 0)
+	one := f.Entry.NewValue0I(f.Entry.Line, OpConst32, f.Config.fe.TypeInt32(), 1)
+
+	// Rewrite backedges to include reschedule checks.
+	for _, emc := range tofixBackedges {
+		e := emc.e
+		headerMemPhi := emc.m
+		headerCtrPhi := emc.c
+		h := e.b
+		i := e.i
+		p := h.Preds[i]
+		bb := p.b
+		mem0 := headerMemPhi.Args[i]
+		ctr0 := headerCtrPhi.Args[i]
+		// bb e->p h,
+		// Because we're going to insert a rare-call, make sure the
+		// looping edge still looks likely.
+		likely := BranchLikely
+		if p.i != 0 {
+			likely = BranchUnlikely
+		}
+		bb.Likely = likely
+
+		// rewrite edge to include reschedule check
+		// existing edges:
+		//
+		// bb.Succs[p.i] == Edge{h, i}
+		// h.Preds[i] == p == Edge{bb,p.i}
+		//
+		// new block(s):
+		// test:
+		//    ctr1 := ctr0 - 1
+		//    if ctr1 <= 0 { goto sched }
+		//    goto join
+		// sched:
+		//    mem1 := call resched (mem0)
+		//    goto join
+		// join:
+		//    ctr2 := phi(ctr1, counter0) // counter0 is the constant
+		//    mem2 := phi(mem0, mem1)
+		//    goto h
+		//
+		// and correct arg i of headerMemPhi and headerCtrPhi
+		//
+		// EXCEPT: block containing only phi functions is bad
+		// for the register allocator.  Therefore, there is no
+		// join, and instead branches targeting join instead target
+		// the header, and the other phi functions within header are
+		// adjusted for the additional input.
+
+		test := f.NewBlock(BlockIf)
+		sched := f.NewBlock(BlockPlain)
+
+		test.Line = bb.Line
+		sched.Line = bb.Line
+
+		//    ctr1 := ctr0 - 1
+		//    if ctr1 <= 0 { goto sched }
+		//    goto header
+		ctr1 := test.NewValue2(bb.Line, OpSub32, f.Config.fe.TypeInt32(), ctr0, one)
+		cmp := test.NewValue2(bb.Line, OpLeq32, f.Config.fe.TypeBool(), ctr1, zero)
+		test.SetControl(cmp)
+		test.AddEdgeTo(sched) // if true
+		// if false -- rewrite edge to header.
+		// do NOT remove+add, because that will perturb all the other phi functions
+		// as well as messing up other edges to the header.
+		test.Succs = append(test.Succs, Edge{h, i})
+		h.Preds[i] = Edge{test, 1}
+		headerMemPhi.SetArg(i, mem0)
+		headerCtrPhi.SetArg(i, ctr1)
+
+		test.Likely = BranchUnlikely
+
+		// sched:
+		//    mem1 := call resched (mem0)
+		//    goto header
+		resched := f.Config.fe.Syslook("goschedguarded")
+		mem1 := sched.NewValue1A(bb.Line, OpStaticCall, TypeMem, resched, mem0)
+		sched.AddEdgeTo(h)
+		headerMemPhi.AddArg(mem1)
+		headerCtrPhi.AddArg(counter0)
+
+		bb.Succs[p.i] = Edge{test, 0}
+		test.Preds = append(test.Preds, Edge{bb, p.i})
+
+		// Must correct all the other phi functions in the header for new incoming edge.
+		// Except for mem and counter phis, it will be the same value seen on the original
+		// backedge at index i.
+		for _, v := range h.Values {
+			if v.Op == OpPhi && v != headerMemPhi && v != headerCtrPhi {
+				v.AddArg(v.Args[i])
+			}
+		}
+	}
+
+	f.invalidateCFG()
+
+	if f.pass.debug > 2 {
+		sdom = newSparseTree(f, f.Idom())
+		fmt.Printf("after %s = %s\n", f.Name, sdom.treestructure(f.Entry))
+	}
+
+	return
+}
+
+// newPhiFor inserts a new Phi function into b,
+// with all inputs set to v.
+func newPhiFor(b *Block, v *Value) *Value {
+	phiV := b.NewValue0(b.Line, OpPhi, v.Type)
+
+	for range b.Preds {
+		phiV.AddArg(v)
+	}
+	return phiV
+}
+
+// rewriteNewPhis updates newphis[h] to record all places where the new phi function inserted
+// in block h will replace a previous definition.  Block b is the block currently being processed;
+// if b has its own phi definition then it takes the place of h.
+// defsForUses provides information about other definitions of the variable that are present
+// (and if nil, indicates that the variable is no longer live)
+func rewriteNewPhis(h, b *Block, f *Func, defsForUses []*Value, newphis map[*Block]rewrite) {
+	// If b is a block with a new phi, then a new rewrite applies below it in the dominator tree.
+	if _, ok := newphis[b]; ok {
+		h = b
+	}
+	change := newphis[h]
+	x := change.before
+	y := change.after
+
+	// Apply rewrites to this block
+	if x != nil { // don't waste time on the common case of no definition.
+		p := &change.rewrites
+		for _, v := range b.Values {
+			if v == y { // don't rewrite self -- phi inputs are handled below.
+				continue
+			}
+			for i, w := range v.Args {
+				if w != x {
+					continue
+				}
+				*p = append(*p, rewriteTarget{v, i})
+			}
+		}
+
+		// Rewrite appropriate inputs of phis reached in successors
+		// in dominance frontier, self, and dominated.
+		// If the variable def reaching uses in b is itself defined in b, then the new phi function
+		// does not reach the successors of b.  (This assumes a bit about the structure of the
+		// phi use-def graph, but it's true for memory and the inserted counter.)
+		if dfu := defsForUses[b.ID]; dfu != nil && dfu.Block != b {
+			for _, e := range b.Succs {
+				s := e.b
+				if sphi, ok := newphis[s]; ok { // saves time to find the phi this way.
+					*p = append(*p, rewriteTarget{sphi.after, e.i})
+					continue
+				}
+				for _, v := range s.Values {
+					if v.Op == OpPhi && v.Args[e.i] == x {
+						*p = append(*p, rewriteTarget{v, e.i})
+						break
+					}
+				}
+			}
+		}
+		newphis[h] = change
+	}
+
+	sdom := f.sdom()
+
+	for c := sdom[b.ID].child; c != nil; c = sdom[c.ID].sibling {
+		rewriteNewPhis(h, c, f, defsForUses, newphis) // TODO: convert to explicit stack from recursion.
+	}
+}
+
+// addDFphis creates new trivial phis that are necessary to correctly reflect (within SSA)
+// a new definition for variable "x" inserted at h (usually but not necessarily a phi).
+// These new phis can only occur at the dominance frontier of h; block s is in the dominance
+// frontier of h if h does not strictly dominate s and if s is a successor of a block b where
+// either b = h or h strictly dominates b.
+// These newly created phis are themselves new definitions that may require addition of their
+// own trivial phi functions in their own dominance frontier, and this is handled recursively.
+func addDFphis(x *Value, h, b *Block, f *Func, defForUses []*Value, newphis map[*Block]rewrite) {
+	oldv := defForUses[b.ID]
+	if oldv != x { // either a new definition replacing x, or nil if it is proven that there are no uses reachable from b
+		return
+	}
+	sdom := f.sdom()
+	idom := f.Idom()
+outer:
+	for _, e := range b.Succs {
+		s := e.b
+		// check phi functions in the dominance frontier
+		if sdom.isAncestor(h, s) {
+			continue // h dominates s, successor of b, therefore s is not in the frontier.
+		}
+		if _, ok := newphis[s]; ok {
+			continue // successor s of b already has a new phi function, so there is no need to add another.
+		}
+		if x != nil {
+			for _, v := range s.Values {
+				if v.Op == OpPhi && v.Args[e.i] == x {
+					continue outer // successor s of b has an old phi function, so there is no need to add another.
+				}
+			}
+		}
+
+		old := defForUses[idom[s.ID].ID] // new phi function is correct-but-redundant, combining value "old" on all inputs.
+		headerPhi := newPhiFor(s, old)
+		// the new phi will replace "old" in block s and all blocks dominated by s.
+		newphis[s] = rewrite{before: old, after: headerPhi} // record new phi, to have inputs labeled "old" rewritten to "headerPhi"
+		addDFphis(old, s, s, f, defForUses, newphis)        // the new definition may also create new phi functions.
+	}
+	for c := sdom[b.ID].child; c != nil; c = sdom[c.ID].sibling {
+		addDFphis(x, h, c, f, defForUses, newphis) // TODO: convert to explicit stack from recursion.
+	}
+}
+
+// findLastMems maps block ids to last memory-output op in a block, if any
+func findLastMems(f *Func) []*Value {
+
+	var stores []*Value
+	lastMems := make([]*Value, f.NumBlocks())
+	storeUse := f.newSparseSet(f.NumValues())
+	defer f.retSparseSet(storeUse)
+	for _, b := range f.Blocks {
+		// Find all the stores in this block. Categorize their uses:
+		//  storeUse contains stores which are used by a subsequent store.
+		storeUse.clear()
+		stores = stores[:0]
+		var memPhi *Value
+		for _, v := range b.Values {
+			if v.Op == OpPhi {
+				if v.Type.IsMemory() {
+					memPhi = v
+				}
+				continue
+			}
+			if v.Type.IsMemory() {
+				stores = append(stores, v)
+				if v.Op == OpSelect1 {
+					// Use the arg of the tuple-generating op.
+					v = v.Args[0]
+				}
+				for _, a := range v.Args {
+					if a.Block == b && a.Type.IsMemory() {
+						storeUse.add(a.ID)
+					}
+				}
+			}
+		}
+		if len(stores) == 0 {
+			lastMems[b.ID] = memPhi
+			continue
+		}
+
+		// find last store in the block
+		var last *Value
+		for _, v := range stores {
+			if storeUse.contains(v.ID) {
+				continue
+			}
+			if last != nil {
+				b.Fatalf("two final stores - simultaneous live stores %s %s", last, v)
+			}
+			last = v
+		}
+		if last == nil {
+			b.Fatalf("no last store found - cycle?")
+		}
+		lastMems[b.ID] = last
+	}
+	return lastMems
+}
+
+type backedgesState struct {
+	b *Block
+	i int
+}
+
+// backedges returns a slice of successor edges that are back
+// edges.  For reducible loops, edge.b is the header.
+func backedges(f *Func) []Edge {
+	edges := []Edge{}
+	mark := make([]markKind, f.NumBlocks())
+	stack := []backedgesState{}
+
+	mark[f.Entry.ID] = notExplored
+	stack = append(stack, backedgesState{f.Entry, 0})
+
+	for len(stack) > 0 {
+		l := len(stack)
+		x := stack[l-1]
+		if x.i < len(x.b.Succs) {
+			e := x.b.Succs[x.i]
+			stack[l-1].i++
+			s := e.b
+			if mark[s.ID] == notFound {
+				mark[s.ID] = notExplored
+				stack = append(stack, backedgesState{s, 0})
+			} else if mark[s.ID] == notExplored {
+				edges = append(edges, e)
+			}
+		} else {
+			mark[x.b.ID] = done
+			stack = stack[0 : l-1]
+		}
+	}
+	return edges
+}

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

@@ -82,7 +82,7 @@ func nilcheckelim(f *Func) {
 				}
 			}
 
-			// Next, process values in the block.
+			// Next, eliminate any redundant nil checks in this block.
 			i := 0
 			for _, v := range b.Values {
 				b.Values[i] = v
@@ -105,13 +105,10 @@ func nilcheckelim(f *Func) {
 							f.Config.Warnl(v.Line, "removed nil check")
 						}
 						v.reset(OpUnknown)
+						// TODO: f.freeValue(v)
 						i--
 						continue
 					}
-					// Record the fact that we know ptr is non nil, and remember to
-					// undo that information when this dominator subtree is done.
-					nonNilValues[ptr.ID] = true
-					work = append(work, bp{op: ClearPtr, ptr: ptr})
 				}
 			}
 			for j := i; j < len(b.Values); j++ {
@@ -119,6 +116,21 @@ func nilcheckelim(f *Func) {
 			}
 			b.Values = b.Values[:i]
 
+			// Finally, find redundant nil checks for subsequent blocks.
+			// Note that we can't add these until the loop above is done, as the
+			// values in the block are not ordered in any way when this pass runs.
+			// This was the cause of issue #18725.
+			for _, v := range b.Values {
+				if v.Op != OpNilCheck {
+					continue
+				}
+				ptr := v.Args[0]
+				// Record the fact that we know ptr is non nil, and remember to
+				// undo that information when this dominator subtree is done.
+				nonNilValues[ptr.ID] = true
+				work = append(work, bp{op: ClearPtr, ptr: ptr})
+			}
+
 			// Add all dominated blocks to the work list.
 			for w := sdom[node.block.ID].child; w != nil; w = sdom[w.ID].sibling {
 				work = append(work, bp{op: Work, block: w})

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

@@ -34,6 +34,7 @@ type opInfo struct {
 	faultOnNilArg0    bool // this op will fault if arg0 is nil (and aux encodes a small offset)
 	faultOnNilArg1    bool // this op will fault if arg1 is nil (and aux encodes a small offset)
 	usesScratch       bool // this op requires scratch memory space
+	hasSideEffects    bool // for "reasons", not to be eliminated.  E.g., atomic store, #19182.
 }
 
 type inputInfo struct {

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

@@ -1473,6 +1473,8 @@ const (
 	OpS390XMOVHZreg
 	OpS390XMOVWreg
 	OpS390XMOVWZreg
+	OpS390XMOVDreg
+	OpS390XMOVDnop
 	OpS390XMOVDconst
 	OpS390XCFDBRA
 	OpS390XCGDBRA
@@ -7305,6 +7307,7 @@ var opcodeTable = [...]opInfo{
 		argLen:         3,
 		resultInArg0:   true,
 		faultOnNilArg1: true,
+		hasSideEffects: true,
 		asm:            x86.AXCHGL,
 		reg: regInfo{
 			inputs: []inputInfo{
@@ -7322,6 +7325,7 @@ var opcodeTable = [...]opInfo{
 		argLen:         3,
 		resultInArg0:   true,
 		faultOnNilArg1: true,
+		hasSideEffects: true,
 		asm:            x86.AXCHGQ,
 		reg: regInfo{
 			inputs: []inputInfo{
@@ -7340,6 +7344,7 @@ var opcodeTable = [...]opInfo{
 		resultInArg0:   true,
 		clobberFlags:   true,
 		faultOnNilArg1: true,
+		hasSideEffects: true,
 		asm:            x86.AXADDL,
 		reg: regInfo{
 			inputs: []inputInfo{
@@ -7358,6 +7363,7 @@ var opcodeTable = [...]opInfo{
 		resultInArg0:   true,
 		clobberFlags:   true,
 		faultOnNilArg1: true,
+		hasSideEffects: true,
 		asm:            x86.AXADDQ,
 		reg: regInfo{
 			inputs: []inputInfo{
@@ -7385,6 +7391,7 @@ var opcodeTable = [...]opInfo{
 		argLen:         4,
 		clobberFlags:   true,
 		faultOnNilArg0: true,
+		hasSideEffects: true,
 		asm:            x86.ACMPXCHGL,
 		reg: regInfo{
 			inputs: []inputInfo{
@@ -7405,6 +7412,7 @@ var opcodeTable = [...]opInfo{
 		argLen:         4,
 		clobberFlags:   true,
 		faultOnNilArg0: true,
+		hasSideEffects: true,
 		asm:            x86.ACMPXCHGQ,
 		reg: regInfo{
 			inputs: []inputInfo{
@@ -7425,6 +7433,7 @@ var opcodeTable = [...]opInfo{
 		argLen:         3,
 		clobberFlags:   true,
 		faultOnNilArg0: true,
+		hasSideEffects: true,
 		asm:            x86.AANDB,
 		reg: regInfo{
 			inputs: []inputInfo{
@@ -7439,6 +7448,7 @@ var opcodeTable = [...]opInfo{
 		argLen:         3,
 		clobberFlags:   true,
 		faultOnNilArg0: true,
+		hasSideEffects: true,
 		asm:            x86.AORB,
 		reg: regInfo{
 			inputs: []inputInfo{
@@ -10073,6 +10083,7 @@ var opcodeTable = [...]opInfo{
 			inputs: []inputInfo{
 				{0, 21503}, // R0 R1 R2 R3 R4 R5 R6 R7 R8 R9 R12 R14
 			},
+			clobbers: 2147483648, // F15
 			outputs: []outputInfo{
 				{0, 4294901760}, // F0 F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 F13 F14 F15
 			},
@@ -10086,6 +10097,7 @@ var opcodeTable = [...]opInfo{
 			inputs: []inputInfo{
 				{0, 21503}, // R0 R1 R2 R3 R4 R5 R6 R7 R8 R9 R12 R14
 			},
+			clobbers: 2147483648, // F15
 			outputs: []outputInfo{
 				{0, 4294901760}, // F0 F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 F13 F14 F15
 			},
@@ -10099,6 +10111,7 @@ var opcodeTable = [...]opInfo{
 			inputs: []inputInfo{
 				{0, 21503}, // R0 R1 R2 R3 R4 R5 R6 R7 R8 R9 R12 R14
 			},
+			clobbers: 2147483648, // F15
 			outputs: []outputInfo{
 				{0, 4294901760}, // F0 F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 F13 F14 F15
 			},
@@ -10112,6 +10125,7 @@ var opcodeTable = [...]opInfo{
 			inputs: []inputInfo{
 				{0, 21503}, // R0 R1 R2 R3 R4 R5 R6 R7 R8 R9 R12 R14
 			},
+			clobbers: 2147483648, // F15
 			outputs: []outputInfo{
 				{0, 4294901760}, // F0 F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 F13 F14 F15
 			},
@@ -10125,6 +10139,7 @@ var opcodeTable = [...]opInfo{
 			inputs: []inputInfo{
 				{0, 4294901760}, // F0 F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 F13 F14 F15
 			},
+			clobbers: 2147483648, // F15
 			outputs: []outputInfo{
 				{0, 21503}, // R0 R1 R2 R3 R4 R5 R6 R7 R8 R9 R12 R14
 			},
@@ -10138,6 +10153,7 @@ var opcodeTable = [...]opInfo{
 			inputs: []inputInfo{
 				{0, 4294901760}, // F0 F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 F13 F14 F15
 			},
+			clobbers: 2147483648, // F15
 			outputs: []outputInfo{
 				{0, 21503}, // R0 R1 R2 R3 R4 R5 R6 R7 R8 R9 R12 R14
 			},
@@ -10151,6 +10167,7 @@ var opcodeTable = [...]opInfo{
 			inputs: []inputInfo{
 				{0, 4294901760}, // F0 F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 F13 F14 F15
 			},
+			clobbers: 2147483648, // F15
 			outputs: []outputInfo{
 				{0, 21503}, // R0 R1 R2 R3 R4 R5 R6 R7 R8 R9 R12 R14
 			},
@@ -10164,6 +10181,7 @@ var opcodeTable = [...]opInfo{
 			inputs: []inputInfo{
 				{0, 4294901760}, // F0 F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 F13 F14 F15
 			},
+			clobbers: 2147483648, // F15
 			outputs: []outputInfo{
 				{0, 21503}, // R0 R1 R2 R3 R4 R5 R6 R7 R8 R9 R12 R14
 			},
@@ -12655,6 +12673,7 @@ var opcodeTable = [...]opInfo{
 		name:           "STLR",
 		argLen:         3,
 		faultOnNilArg0: true,
+		hasSideEffects: true,
 		asm:            arm64.ASTLR,
 		reg: regInfo{
 			inputs: []inputInfo{
@@ -12667,6 +12686,7 @@ var opcodeTable = [...]opInfo{
 		name:           "STLRW",
 		argLen:         3,
 		faultOnNilArg0: true,
+		hasSideEffects: true,
 		asm:            arm64.ASTLRW,
 		reg: regInfo{
 			inputs: []inputInfo{
@@ -12680,6 +12700,7 @@ var opcodeTable = [...]opInfo{
 		argLen:          3,
 		resultNotInArgs: true,
 		faultOnNilArg0:  true,
+		hasSideEffects:  true,
 		reg: regInfo{
 			inputs: []inputInfo{
 				{1, 805044223},           // R0 R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R19 R20 R21 R22 R23 R24 R25 R26 g R30
@@ -12695,6 +12716,7 @@ var opcodeTable = [...]opInfo{
 		argLen:          3,
 		resultNotInArgs: true,
 		faultOnNilArg0:  true,
+		hasSideEffects:  true,
 		reg: regInfo{
 			inputs: []inputInfo{
 				{1, 805044223},           // R0 R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R19 R20 R21 R22 R23 R24 R25 R26 g R30
@@ -12710,6 +12732,7 @@ var opcodeTable = [...]opInfo{
 		argLen:          3,
 		resultNotInArgs: true,
 		faultOnNilArg0:  true,
+		hasSideEffects:  true,
 		reg: regInfo{
 			inputs: []inputInfo{
 				{1, 805044223},           // R0 R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R19 R20 R21 R22 R23 R24 R25 R26 g R30
@@ -12725,6 +12748,7 @@ var opcodeTable = [...]opInfo{
 		argLen:          3,
 		resultNotInArgs: true,
 		faultOnNilArg0:  true,
+		hasSideEffects:  true,
 		reg: regInfo{
 			inputs: []inputInfo{
 				{1, 805044223},           // R0 R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R19 R20 R21 R22 R23 R24 R25 R26 g R30
@@ -12741,6 +12765,7 @@ var opcodeTable = [...]opInfo{
 		resultNotInArgs: true,
 		clobberFlags:    true,
 		faultOnNilArg0:  true,
+		hasSideEffects:  true,
 		reg: regInfo{
 			inputs: []inputInfo{
 				{1, 805044223},           // R0 R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R19 R20 R21 R22 R23 R24 R25 R26 g R30
@@ -12758,6 +12783,7 @@ var opcodeTable = [...]opInfo{
 		resultNotInArgs: true,
 		clobberFlags:    true,
 		faultOnNilArg0:  true,
+		hasSideEffects:  true,
 		reg: regInfo{
 			inputs: []inputInfo{
 				{1, 805044223},           // R0 R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R19 R20 R21 R22 R23 R24 R25 R26 g R30
@@ -12773,6 +12799,7 @@ var opcodeTable = [...]opInfo{
 		name:           "LoweredAtomicAnd8",
 		argLen:         3,
 		faultOnNilArg0: true,
+		hasSideEffects: true,
 		asm:            arm64.AAND,
 		reg: regInfo{
 			inputs: []inputInfo{
@@ -12785,6 +12812,7 @@ var opcodeTable = [...]opInfo{
 		name:           "LoweredAtomicOr8",
 		argLen:         3,
 		faultOnNilArg0: true,
+		hasSideEffects: true,
 		asm:            arm64.AORR,
 		reg: regInfo{
 			inputs: []inputInfo{
@@ -13975,6 +14003,7 @@ var opcodeTable = [...]opInfo{
 		name:           "LoweredAtomicStore",
 		argLen:         3,
 		faultOnNilArg0: true,
+		hasSideEffects: true,
 		reg: regInfo{
 			inputs: []inputInfo{
 				{1, 469762046},       // R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R24 R25 R28 g R31
@@ -13986,6 +14015,7 @@ var opcodeTable = [...]opInfo{
 		name:           "LoweredAtomicStorezero",
 		argLen:         2,
 		faultOnNilArg0: true,
+		hasSideEffects: true,
 		reg: regInfo{
 			inputs: []inputInfo{
 				{0, 140738025226238}, // R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R24 R25 R28 SP g R31 SB
@@ -13997,6 +14027,7 @@ var opcodeTable = [...]opInfo{
 		argLen:          3,
 		resultNotInArgs: true,
 		faultOnNilArg0:  true,
+		hasSideEffects:  true,
 		reg: regInfo{
 			inputs: []inputInfo{
 				{1, 469762046},       // R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R24 R25 R28 g R31
@@ -14012,6 +14043,7 @@ var opcodeTable = [...]opInfo{
 		argLen:          3,
 		resultNotInArgs: true,
 		faultOnNilArg0:  true,
+		hasSideEffects:  true,
 		reg: regInfo{
 			inputs: []inputInfo{
 				{1, 469762046},       // R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R24 R25 R28 g R31
@@ -14028,6 +14060,7 @@ var opcodeTable = [...]opInfo{
 		argLen:          2,
 		resultNotInArgs: true,
 		faultOnNilArg0:  true,
+		hasSideEffects:  true,
 		reg: regInfo{
 			inputs: []inputInfo{
 				{0, 140738025226238}, // R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R24 R25 R28 SP g R31 SB
@@ -14042,6 +14075,7 @@ var opcodeTable = [...]opInfo{
 		argLen:          4,
 		resultNotInArgs: true,
 		faultOnNilArg0:  true,
+		hasSideEffects:  true,
 		reg: regInfo{
 			inputs: []inputInfo{
 				{1, 469762046},       // R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R24 R25 R28 g R31
@@ -14057,6 +14091,7 @@ var opcodeTable = [...]opInfo{
 		name:           "LoweredAtomicAnd",
 		argLen:         3,
 		faultOnNilArg0: true,
+		hasSideEffects: true,
 		asm:            mips.AAND,
 		reg: regInfo{
 			inputs: []inputInfo{
@@ -14069,6 +14104,7 @@ var opcodeTable = [...]opInfo{
 		name:           "LoweredAtomicOr",
 		argLen:         3,
 		faultOnNilArg0: true,
+		hasSideEffects: true,
 		asm:            mips.AOR,
 		reg: regInfo{
 			inputs: []inputInfo{
@@ -18570,6 +18606,32 @@ var opcodeTable = [...]opInfo{
 			},
 		},
 	},
+	{
+		name:   "MOVDreg",
+		argLen: 1,
+		asm:    s390x.AMOVD,
+		reg: regInfo{
+			inputs: []inputInfo{
+				{0, 54271}, // R0 R1 R2 R3 R4 R5 R6 R7 R8 R9 R12 R14 SP
+			},
+			outputs: []outputInfo{
+				{0, 21503}, // R0 R1 R2 R3 R4 R5 R6 R7 R8 R9 R12 R14
+			},
+		},
+	},
+	{
+		name:         "MOVDnop",
+		argLen:       1,
+		resultInArg0: true,
+		reg: regInfo{
+			inputs: []inputInfo{
+				{0, 21503}, // R0 R1 R2 R3 R4 R5 R6 R7 R8 R9 R12 R14
+			},
+			outputs: []outputInfo{
+				{0, 21503}, // R0 R1 R2 R3 R4 R5 R6 R7 R8 R9 R12 R14
+			},
+		},
+	},
 	{
 		name:              "MOVDconst",
 		auxType:           auxInt64,
@@ -19472,6 +19534,7 @@ var opcodeTable = [...]opInfo{
 		argLen:         3,
 		clobberFlags:   true,
 		faultOnNilArg0: true,
+		hasSideEffects: true,
 		asm:            s390x.AMOVW,
 		reg: regInfo{
 			inputs: []inputInfo{
@@ -19486,6 +19549,7 @@ var opcodeTable = [...]opInfo{
 		argLen:         3,
 		clobberFlags:   true,
 		faultOnNilArg0: true,
+		hasSideEffects: true,
 		asm:            s390x.AMOVD,
 		reg: regInfo{
 			inputs: []inputInfo{
@@ -19499,6 +19563,7 @@ var opcodeTable = [...]opInfo{
 		auxType:        auxSymOff,
 		argLen:         3,
 		faultOnNilArg0: true,
+		hasSideEffects: true,
 		asm:            s390x.ALAA,
 		reg: regInfo{
 			inputs: []inputInfo{
@@ -19515,6 +19580,7 @@ var opcodeTable = [...]opInfo{
 		auxType:        auxSymOff,
 		argLen:         3,
 		faultOnNilArg0: true,
+		hasSideEffects: true,
 		asm:            s390x.ALAAG,
 		reg: regInfo{
 			inputs: []inputInfo{
@@ -19542,6 +19608,7 @@ var opcodeTable = [...]opInfo{
 		argLen:         4,
 		clobberFlags:   true,
 		faultOnNilArg0: true,
+		hasSideEffects: true,
 		asm:            s390x.ACS,
 		reg: regInfo{
 			inputs: []inputInfo{
@@ -19562,6 +19629,7 @@ var opcodeTable = [...]opInfo{
 		argLen:         4,
 		clobberFlags:   true,
 		faultOnNilArg0: true,
+		hasSideEffects: true,
 		asm:            s390x.ACSG,
 		reg: regInfo{
 			inputs: []inputInfo{
@@ -19582,6 +19650,7 @@ var opcodeTable = [...]opInfo{
 		argLen:         3,
 		clobberFlags:   true,
 		faultOnNilArg0: true,
+		hasSideEffects: true,
 		asm:            s390x.ACS,
 		reg: regInfo{
 			inputs: []inputInfo{
@@ -19600,6 +19669,7 @@ var opcodeTable = [...]opInfo{
 		argLen:         3,
 		clobberFlags:   true,
 		faultOnNilArg0: true,
+		hasSideEffects: true,
 		asm:            s390x.ACSG,
 		reg: regInfo{
 			inputs: []inputInfo{
@@ -21387,59 +21457,70 @@ var opcodeTable = [...]opInfo{
 		generic: true,
 	},
 	{
-		name:    "AtomicStore32",
-		argLen:  3,
-		generic: true,
+		name:           "AtomicStore32",
+		argLen:         3,
+		hasSideEffects: true,
+		generic:        true,
 	},
 	{
-		name:    "AtomicStore64",
-		argLen:  3,
-		generic: true,
+		name:           "AtomicStore64",
+		argLen:         3,
+		hasSideEffects: true,
+		generic:        true,
 	},
 	{
-		name:    "AtomicStorePtrNoWB",
-		argLen:  3,
-		generic: true,
+		name:           "AtomicStorePtrNoWB",
+		argLen:         3,
+		hasSideEffects: true,
+		generic:        true,
 	},
 	{
-		name:    "AtomicExchange32",
-		argLen:  3,
-		generic: true,
+		name:           "AtomicExchange32",
+		argLen:         3,
+		hasSideEffects: true,
+		generic:        true,
 	},
 	{
-		name:    "AtomicExchange64",
-		argLen:  3,
-		generic: true,
+		name:           "AtomicExchange64",
+		argLen:         3,
+		hasSideEffects: true,
+		generic:        true,
 	},
 	{
-		name:    "AtomicAdd32",
-		argLen:  3,
-		generic: true,
+		name:           "AtomicAdd32",
+		argLen:         3,
+		hasSideEffects: true,
+		generic:        true,
 	},
 	{
-		name:    "AtomicAdd64",
-		argLen:  3,
-		generic: true,
+		name:           "AtomicAdd64",
+		argLen:         3,
+		hasSideEffects: true,
+		generic:        true,
 	},
 	{
-		name:    "AtomicCompareAndSwap32",
-		argLen:  4,
-		generic: true,
+		name:           "AtomicCompareAndSwap32",
+		argLen:         4,
+		hasSideEffects: true,
+		generic:        true,
 	},
 	{
-		name:    "AtomicCompareAndSwap64",
-		argLen:  4,
-		generic: true,
+		name:           "AtomicCompareAndSwap64",
+		argLen:         4,
+		hasSideEffects: true,
+		generic:        true,
 	},
 	{
-		name:    "AtomicAnd8",
-		argLen:  3,
-		generic: true,
+		name:           "AtomicAnd8",
+		argLen:         3,
+		hasSideEffects: true,
+		generic:        true,
 	},
 	{
-		name:    "AtomicOr8",
-		argLen:  3,
-		generic: true,
+		name:           "AtomicOr8",
+		argLen:         3,
+		hasSideEffects: true,
+		generic:        true,
 	},
 }
 

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

@@ -1699,6 +1699,24 @@ sinking:
 			}
 			p := d.Preds[0].b // block in loop exiting to d.
 
+			// Check that the spill value is still live at the start of d.
+			// If it isn't, we can't move the spill here because some other value
+			// may be using the same stack slot.  See issue 20472.
+			// The spill value can't be defined in d, so that makes our lives easier.
+			for _, x := range stacklive[d.ID] {
+				if x == vsp.ID {
+					goto stillLive
+				}
+			}
+			for _, v := range d.Values {
+				if v.Op == OpLoadReg && v.Args[0] == vsp {
+					goto stillLive
+				}
+			}
+			// Spill is not live - abort sinking this spill.
+			continue sinking
+		stillLive:
+
 			endregs := s.endRegs[p.ID]
 			for _, regrec := range endregs {
 				if regrec.v == e && regrec.r != noRegister && regrec.c == e { // TODO: regrec.c != e implies different spill possible.

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

@@ -31,3 +31,68 @@ func TestLiveControlOps(t *testing.T) {
 	regalloc(f.f)
 	checkFunc(f.f)
 }
+
+func TestSpillMove(t *testing.T) {
+	// Test for issue 20472. We shouldn't move a spill out to exit blocks
+	// if there is an exit block where the spill is dead but the pre-spill
+	// value is live.
+	c := testConfig(t)
+	ptrType := &TypeImpl{Size_: 8, Ptr: true, Name: "testptr"} // dummy for testing
+	arg1Aux := c.fe.Auto(TypeInt64)
+	arg2Aux := c.fe.Auto(ptrType)
+	f := Fun(c, "entry",
+		Bloc("entry",
+			Valu("mem", OpInitMem, TypeMem, 0, nil),
+			Valu("x", OpArg, TypeInt64, 0, arg1Aux),
+			Valu("p", OpArg, ptrType, 0, arg2Aux),
+			Valu("a", OpAMD64TESTQ, TypeFlags, 0, nil, "x", "x"),
+			Goto("loop1"),
+		),
+		Bloc("loop1",
+			Valu("y", OpAMD64MULQ, TypeInt64, 0, nil, "x", "x"),
+			Eq("a", "loop2", "exit1"),
+		),
+		Bloc("loop2",
+			Eq("a", "loop1", "exit2"),
+		),
+		Bloc("exit1",
+			// store before call, y is available in a register
+			Valu("mem2", OpAMD64MOVQstore, TypeMem, 0, nil, "p", "y", "mem"),
+			Valu("mem3", OpAMD64CALLstatic, TypeMem, 0, nil, "mem2"),
+			Exit("mem3"),
+		),
+		Bloc("exit2",
+			// store after call, y must be loaded from a spill location
+			Valu("mem4", OpAMD64CALLstatic, TypeMem, 0, nil, "mem"),
+			Valu("mem5", OpAMD64MOVQstore, TypeMem, 0, nil, "p", "y", "mem4"),
+			Exit("mem5"),
+		),
+	)
+	flagalloc(f.f)
+	regalloc(f.f)
+	checkFunc(f.f)
+	// There should still be a spill in Loop1, and nowhere else.
+	if numSpills(f.blocks["loop1"]) != 1 {
+		t.Errorf("spill missing from loop1")
+	}
+	if numSpills(f.blocks["loop2"]) != 0 {
+		t.Errorf("spill present in loop2")
+	}
+	if numSpills(f.blocks["exit1"]) != 0 {
+		t.Errorf("spill present in exit1")
+	}
+	if numSpills(f.blocks["exit2"]) != 0 {
+		t.Errorf("spill present in exit2")
+	}
+
+}
+
+func numSpills(b *Block) int {
+	n := 0
+	for _, v := range b.Values {
+		if v.Op == OpStoreReg {
+			n++
+		}
+	}
+	return n
+}

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

@@ -14517,7 +14517,7 @@ func rewriteValueARM_OpLrot16(v *Value, config *Config) bool {
 	_ = b
 	// match: (Lrot16 <t> x [c])
 	// cond:
-	// result: (OR (SLLconst <t> x [c&15]) (SRLconst <t> x [16-c&15]))
+	// result: (OR (SLLconst <t> x [c&15]) (SRLconst <t> (ZeroExt16to32 x) [16-c&15]))
 	for {
 		t := v.Type
 		c := v.AuxInt
@@ -14529,7 +14529,9 @@ func rewriteValueARM_OpLrot16(v *Value, config *Config) bool {
 		v.AddArg(v0)
 		v1 := b.NewValue0(v.Line, OpARMSRLconst, t)
 		v1.AuxInt = 16 - c&15
-		v1.AddArg(x)
+		v2 := b.NewValue0(v.Line, OpZeroExt16to32, config.fe.TypeUInt32())
+		v2.AddArg(x)
+		v1.AddArg(v2)
 		v.AddArg(v1)
 		return true
 	}
@@ -14554,7 +14556,7 @@ func rewriteValueARM_OpLrot8(v *Value, config *Config) bool {
 	_ = b
 	// match: (Lrot8 <t> x [c])
 	// cond:
-	// result: (OR (SLLconst <t> x [c&7]) (SRLconst <t> x [8-c&7]))
+	// result: (OR (SLLconst <t> x [c&7]) (SRLconst <t> (ZeroExt8to32 x) [8-c&7]))
 	for {
 		t := v.Type
 		c := v.AuxInt
@@ -14566,7 +14568,9 @@ func rewriteValueARM_OpLrot8(v *Value, config *Config) bool {
 		v.AddArg(v0)
 		v1 := b.NewValue0(v.Line, OpARMSRLconst, t)
 		v1.AuxInt = 8 - c&7
-		v1.AddArg(x)
+		v2 := b.NewValue0(v.Line, OpZeroExt8to32, config.fe.TypeUInt32())
+		v2.AddArg(x)
+		v1.AddArg(v2)
 		v.AddArg(v1)
 		return true
 	}

src/cmd/compile/internal/ssa/rewriteARM64.go

@@ -2625,7 +2625,7 @@ func rewriteValueARM64_OpARM64FMOVDload(v *Value, config *Config) bool {
 	b := v.Block
 	_ = b
 	// match: (FMOVDload [off1] {sym} (ADDconst [off2] ptr) mem)
-	// cond: (off1+off2)%8==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym) && !isAuto(sym)
+	// cond: is32Bit(off1+off2) && !isArg(sym) 	&& ((off1+off2)%8==0 || off1+off2<256 && off1+off2>-256 && !isAuto(sym))
 	// result: (FMOVDload [off1+off2] {sym} ptr mem)
 	for {
 		off1 := v.AuxInt
@@ -2637,7 +2637,7 @@ func rewriteValueARM64_OpARM64FMOVDload(v *Value, config *Config) bool {
 		off2 := v_0.AuxInt
 		ptr := v_0.Args[0]
 		mem := v.Args[1]
-		if !((off1+off2)%8 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isArg(sym) && !isAuto(sym)) {
+		if !(is32Bit(off1+off2) && !isArg(sym) && ((off1+off2)%8 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isAuto(sym))) {
 			break
 		}
 		v.reset(OpARM64FMOVDload)
@@ -2648,7 +2648,7 @@ func rewriteValueARM64_OpARM64FMOVDload(v *Value, config *Config) bool {
 		return true
 	}
 	// match: (FMOVDload [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem)
-	// cond: canMergeSym(sym1,sym2) 	&& ((off1+off2)%8==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym1) && !isAuto(sym1))
+	// cond: canMergeSym(sym1,sym2) 	&& is32Bit(off1+off2) && !isArg(mergeSym(sym1,sym2)) 	&& ((off1+off2)%8==0 || off1+off2<256 && off1+off2>-256 && !isAuto(mergeSym(sym1,sym2)))
 	// result: (FMOVDload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
 	for {
 		off1 := v.AuxInt
@@ -2661,7 +2661,7 @@ func rewriteValueARM64_OpARM64FMOVDload(v *Value, config *Config) bool {
 		sym2 := v_0.Aux
 		ptr := v_0.Args[0]
 		mem := v.Args[1]
-		if !(canMergeSym(sym1, sym2) && ((off1+off2)%8 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isArg(sym1) && !isAuto(sym1))) {
+		if !(canMergeSym(sym1, sym2) && is32Bit(off1+off2) && !isArg(mergeSym(sym1, sym2)) && ((off1+off2)%8 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isAuto(mergeSym(sym1, sym2)))) {
 			break
 		}
 		v.reset(OpARM64FMOVDload)
@@ -2677,7 +2677,7 @@ func rewriteValueARM64_OpARM64FMOVDstore(v *Value, config *Config) bool {
 	b := v.Block
 	_ = b
 	// match: (FMOVDstore [off1] {sym} (ADDconst [off2] ptr) val mem)
-	// cond: (off1+off2)%8==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym) && !isAuto(sym)
+	// cond: is32Bit(off1+off2) && !isArg(sym) 	&& ((off1+off2)%8==0 || off1+off2<256 && off1+off2>-256 && !isAuto(sym))
 	// result: (FMOVDstore [off1+off2] {sym} ptr val mem)
 	for {
 		off1 := v.AuxInt
@@ -2690,7 +2690,7 @@ func rewriteValueARM64_OpARM64FMOVDstore(v *Value, config *Config) bool {
 		ptr := v_0.Args[0]
 		val := v.Args[1]
 		mem := v.Args[2]
-		if !((off1+off2)%8 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isArg(sym) && !isAuto(sym)) {
+		if !(is32Bit(off1+off2) && !isArg(sym) && ((off1+off2)%8 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isAuto(sym))) {
 			break
 		}
 		v.reset(OpARM64FMOVDstore)
@@ -2702,7 +2702,7 @@ func rewriteValueARM64_OpARM64FMOVDstore(v *Value, config *Config) bool {
 		return true
 	}
 	// match: (FMOVDstore [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) val mem)
-	// cond: canMergeSym(sym1,sym2) 	&& ((off1+off2)%8==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym1) && !isAuto(sym1))
+	// cond: canMergeSym(sym1,sym2) 	&& is32Bit(off1+off2) && !isArg(mergeSym(sym1,sym2)) 	&& ((off1+off2)%8==0 || off1+off2<256 && off1+off2>-256 && !isAuto(mergeSym(sym1,sym2)))
 	// result: (FMOVDstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem)
 	for {
 		off1 := v.AuxInt
@@ -2716,7 +2716,7 @@ func rewriteValueARM64_OpARM64FMOVDstore(v *Value, config *Config) bool {
 		ptr := v_0.Args[0]
 		val := v.Args[1]
 		mem := v.Args[2]
-		if !(canMergeSym(sym1, sym2) && ((off1+off2)%8 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isArg(sym1) && !isAuto(sym1))) {
+		if !(canMergeSym(sym1, sym2) && is32Bit(off1+off2) && !isArg(mergeSym(sym1, sym2)) && ((off1+off2)%8 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isAuto(mergeSym(sym1, sym2)))) {
 			break
 		}
 		v.reset(OpARM64FMOVDstore)
@@ -2733,7 +2733,7 @@ func rewriteValueARM64_OpARM64FMOVSload(v *Value, config *Config) bool {
 	b := v.Block
 	_ = b
 	// match: (FMOVSload [off1] {sym} (ADDconst [off2] ptr) mem)
-	// cond: (off1+off2)%4==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym) && !isAuto(sym)
+	// cond: is32Bit(off1+off2) && !isArg(sym) 	&& ((off1+off2)%4==0 || off1+off2<256 && off1+off2>-256 && !isAuto(sym))
 	// result: (FMOVSload [off1+off2] {sym} ptr mem)
 	for {
 		off1 := v.AuxInt
@@ -2745,7 +2745,7 @@ func rewriteValueARM64_OpARM64FMOVSload(v *Value, config *Config) bool {
 		off2 := v_0.AuxInt
 		ptr := v_0.Args[0]
 		mem := v.Args[1]
-		if !((off1+off2)%4 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isArg(sym) && !isAuto(sym)) {
+		if !(is32Bit(off1+off2) && !isArg(sym) && ((off1+off2)%4 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isAuto(sym))) {
 			break
 		}
 		v.reset(OpARM64FMOVSload)
@@ -2756,7 +2756,7 @@ func rewriteValueARM64_OpARM64FMOVSload(v *Value, config *Config) bool {
 		return true
 	}
 	// match: (FMOVSload [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem)
-	// cond: canMergeSym(sym1,sym2) 	&& ((off1+off2)%4==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym1) && !isAuto(sym1))
+	// cond: canMergeSym(sym1,sym2) 	&& is32Bit(off1+off2) && !isArg(mergeSym(sym1,sym2)) 	&& ((off1+off2)%4==0 || off1+off2<256 && off1+off2>-256 && !isAuto(mergeSym(sym1,sym2)))
 	// result: (FMOVSload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
 	for {
 		off1 := v.AuxInt
@@ -2769,7 +2769,7 @@ func rewriteValueARM64_OpARM64FMOVSload(v *Value, config *Config) bool {
 		sym2 := v_0.Aux
 		ptr := v_0.Args[0]
 		mem := v.Args[1]
-		if !(canMergeSym(sym1, sym2) && ((off1+off2)%4 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isArg(sym1) && !isAuto(sym1))) {
+		if !(canMergeSym(sym1, sym2) && is32Bit(off1+off2) && !isArg(mergeSym(sym1, sym2)) && ((off1+off2)%4 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isAuto(mergeSym(sym1, sym2)))) {
 			break
 		}
 		v.reset(OpARM64FMOVSload)
@@ -2785,7 +2785,7 @@ func rewriteValueARM64_OpARM64FMOVSstore(v *Value, config *Config) bool {
 	b := v.Block
 	_ = b
 	// match: (FMOVSstore [off1] {sym} (ADDconst [off2] ptr) val mem)
-	// cond: (off1+off2)%4==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym) && !isAuto(sym)
+	// cond: is32Bit(off1+off2) && !isArg(sym) 	&& ((off1+off2)%4==0 || off1+off2<256 && off1+off2>-256 && !isAuto(sym))
 	// result: (FMOVSstore [off1+off2] {sym} ptr val mem)
 	for {
 		off1 := v.AuxInt
@@ -2798,7 +2798,7 @@ func rewriteValueARM64_OpARM64FMOVSstore(v *Value, config *Config) bool {
 		ptr := v_0.Args[0]
 		val := v.Args[1]
 		mem := v.Args[2]
-		if !((off1+off2)%4 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isArg(sym) && !isAuto(sym)) {
+		if !(is32Bit(off1+off2) && !isArg(sym) && ((off1+off2)%4 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isAuto(sym))) {
 			break
 		}
 		v.reset(OpARM64FMOVSstore)
@@ -2810,7 +2810,7 @@ func rewriteValueARM64_OpARM64FMOVSstore(v *Value, config *Config) bool {
 		return true
 	}
 	// match: (FMOVSstore [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) val mem)
-	// cond: canMergeSym(sym1,sym2) 	&& ((off1+off2)%4==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym1) && !isAuto(sym1))
+	// cond: canMergeSym(sym1,sym2) 	&& is32Bit(off1+off2) && !isArg(mergeSym(sym1,sym2)) 	&& ((off1+off2)%4==0 || off1+off2<256 && off1+off2>-256 && !isAuto(mergeSym(sym1,sym2)))
 	// result: (FMOVSstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem)
 	for {
 		off1 := v.AuxInt
@@ -2824,7 +2824,7 @@ func rewriteValueARM64_OpARM64FMOVSstore(v *Value, config *Config) bool {
 		ptr := v_0.Args[0]
 		val := v.Args[1]
 		mem := v.Args[2]
-		if !(canMergeSym(sym1, sym2) && ((off1+off2)%4 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isArg(sym1) && !isAuto(sym1))) {
+		if !(canMergeSym(sym1, sym2) && is32Bit(off1+off2) && !isArg(mergeSym(sym1, sym2)) && ((off1+off2)%4 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isAuto(mergeSym(sym1, sym2)))) {
 			break
 		}
 		v.reset(OpARM64FMOVSstore)
@@ -3511,7 +3511,7 @@ func rewriteValueARM64_OpARM64MOVBUload(v *Value, config *Config) bool {
 	b := v.Block
 	_ = b
 	// match: (MOVBUload [off1] {sym} (ADDconst [off2] ptr) mem)
-	// cond:
+	// cond: is32Bit(off1+off2)
 	// result: (MOVBUload [off1+off2] {sym} ptr mem)
 	for {
 		off1 := v.AuxInt
@@ -3523,6 +3523,9 @@ func rewriteValueARM64_OpARM64MOVBUload(v *Value, config *Config) bool {
 		off2 := v_0.AuxInt
 		ptr := v_0.Args[0]
 		mem := v.Args[1]
+		if !(is32Bit(off1 + off2)) {
+			break
+		}
 		v.reset(OpARM64MOVBUload)
 		v.AuxInt = off1 + off2
 		v.Aux = sym
@@ -3531,7 +3534,7 @@ func rewriteValueARM64_OpARM64MOVBUload(v *Value, config *Config) bool {
 		return true
 	}
 	// match: (MOVBUload [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem)
-	// cond: canMergeSym(sym1,sym2)
+	// cond: canMergeSym(sym1,sym2) 	&& is32Bit(off1+off2)
 	// result: (MOVBUload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
 	for {
 		off1 := v.AuxInt
@@ -3544,7 +3547,7 @@ func rewriteValueARM64_OpARM64MOVBUload(v *Value, config *Config) bool {
 		sym2 := v_0.Aux
 		ptr := v_0.Args[0]
 		mem := v.Args[1]
-		if !(canMergeSym(sym1, sym2)) {
+		if !(canMergeSym(sym1, sym2) && is32Bit(off1+off2)) {
 			break
 		}
 		v.reset(OpARM64MOVBUload)
@@ -3623,7 +3626,7 @@ func rewriteValueARM64_OpARM64MOVBload(v *Value, config *Config) bool {
 	b := v.Block
 	_ = b
 	// match: (MOVBload [off1] {sym} (ADDconst [off2] ptr) mem)
-	// cond:
+	// cond: is32Bit(off1+off2)
 	// result: (MOVBload [off1+off2] {sym} ptr mem)
 	for {
 		off1 := v.AuxInt
@@ -3635,6 +3638,9 @@ func rewriteValueARM64_OpARM64MOVBload(v *Value, config *Config) bool {
 		off2 := v_0.AuxInt
 		ptr := v_0.Args[0]
 		mem := v.Args[1]
+		if !(is32Bit(off1 + off2)) {
+			break
+		}
 		v.reset(OpARM64MOVBload)
 		v.AuxInt = off1 + off2
 		v.Aux = sym
@@ -3643,7 +3649,7 @@ func rewriteValueARM64_OpARM64MOVBload(v *Value, config *Config) bool {
 		return true
 	}
 	// match: (MOVBload [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem)
-	// cond: canMergeSym(sym1,sym2)
+	// cond: canMergeSym(sym1,sym2) 	&& is32Bit(off1+off2)
 	// result: (MOVBload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
 	for {
 		off1 := v.AuxInt
@@ -3656,7 +3662,7 @@ func rewriteValueARM64_OpARM64MOVBload(v *Value, config *Config) bool {
 		sym2 := v_0.Aux
 		ptr := v_0.Args[0]
 		mem := v.Args[1]
-		if !(canMergeSym(sym1, sym2)) {
+		if !(canMergeSym(sym1, sym2) && is32Bit(off1+off2)) {
 			break
 		}
 		v.reset(OpARM64MOVBload)
@@ -3735,7 +3741,7 @@ func rewriteValueARM64_OpARM64MOVBstore(v *Value, config *Config) bool {
 	b := v.Block
 	_ = b
 	// match: (MOVBstore [off1] {sym} (ADDconst [off2] ptr) val mem)
-	// cond:
+	// cond: is32Bit(off1+off2)
 	// result: (MOVBstore [off1+off2] {sym} ptr val mem)
 	for {
 		off1 := v.AuxInt
@@ -3748,6 +3754,9 @@ func rewriteValueARM64_OpARM64MOVBstore(v *Value, config *Config) bool {
 		ptr := v_0.Args[0]
 		val := v.Args[1]
 		mem := v.Args[2]
+		if !(is32Bit(off1 + off2)) {
+			break
+		}
 		v.reset(OpARM64MOVBstore)
 		v.AuxInt = off1 + off2
 		v.Aux = sym
@@ -3757,7 +3766,7 @@ func rewriteValueARM64_OpARM64MOVBstore(v *Value, config *Config) bool {
 		return true
 	}
 	// match: (MOVBstore [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) val mem)
-	// cond: canMergeSym(sym1,sym2)
+	// cond: canMergeSym(sym1,sym2) 	&& is32Bit(off1+off2)
 	// result: (MOVBstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem)
 	for {
 		off1 := v.AuxInt
@@ -3771,7 +3780,7 @@ func rewriteValueARM64_OpARM64MOVBstore(v *Value, config *Config) bool {
 		ptr := v_0.Args[0]
 		val := v.Args[1]
 		mem := v.Args[2]
-		if !(canMergeSym(sym1, sym2)) {
+		if !(canMergeSym(sym1, sym2) && is32Bit(off1+off2)) {
 			break
 		}
 		v.reset(OpARM64MOVBstore)
@@ -3936,7 +3945,7 @@ func rewriteValueARM64_OpARM64MOVBstorezero(v *Value, config *Config) bool {
 	b := v.Block
 	_ = b
 	// match: (MOVBstorezero [off1] {sym} (ADDconst [off2] ptr) mem)
-	// cond:
+	// cond: is32Bit(off1+off2)
 	// result: (MOVBstorezero [off1+off2] {sym} ptr mem)
 	for {
 		off1 := v.AuxInt
@@ -3948,6 +3957,9 @@ func rewriteValueARM64_OpARM64MOVBstorezero(v *Value, config *Config) bool {
 		off2 := v_0.AuxInt
 		ptr := v_0.Args[0]
 		mem := v.Args[1]
+		if !(is32Bit(off1 + off2)) {
+			break
+		}
 		v.reset(OpARM64MOVBstorezero)
 		v.AuxInt = off1 + off2
 		v.Aux = sym
@@ -3956,7 +3968,7 @@ func rewriteValueARM64_OpARM64MOVBstorezero(v *Value, config *Config) bool {
 		return true
 	}
 	// match: (MOVBstorezero [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem)
-	// cond: canMergeSym(sym1,sym2)
+	// cond: canMergeSym(sym1,sym2) 	&& is32Bit(off1+off2)
 	// result: (MOVBstorezero [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
 	for {
 		off1 := v.AuxInt
@@ -3969,7 +3981,7 @@ func rewriteValueARM64_OpARM64MOVBstorezero(v *Value, config *Config) bool {
 		sym2 := v_0.Aux
 		ptr := v_0.Args[0]
 		mem := v.Args[1]
-		if !(canMergeSym(sym1, sym2)) {
+		if !(canMergeSym(sym1, sym2) && is32Bit(off1+off2)) {
 			break
 		}
 		v.reset(OpARM64MOVBstorezero)
@@ -3985,7 +3997,7 @@ func rewriteValueARM64_OpARM64MOVDload(v *Value, config *Config) bool {
 	b := v.Block
 	_ = b
 	// match: (MOVDload [off1] {sym} (ADDconst [off2] ptr) mem)
-	// cond: (off1+off2)%8==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym) && !isAuto(sym)
+	// cond: is32Bit(off1+off2) && !isArg(sym) 	&& ((off1+off2)%8==0 || off1+off2<256 && off1+off2>-256 && !isAuto(sym))
 	// result: (MOVDload [off1+off2] {sym} ptr mem)
 	for {
 		off1 := v.AuxInt
@@ -3997,7 +4009,7 @@ func rewriteValueARM64_OpARM64MOVDload(v *Value, config *Config) bool {
 		off2 := v_0.AuxInt
 		ptr := v_0.Args[0]
 		mem := v.Args[1]
-		if !((off1+off2)%8 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isArg(sym) && !isAuto(sym)) {
+		if !(is32Bit(off1+off2) && !isArg(sym) && ((off1+off2)%8 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isAuto(sym))) {
 			break
 		}
 		v.reset(OpARM64MOVDload)
@@ -4008,7 +4020,7 @@ func rewriteValueARM64_OpARM64MOVDload(v *Value, config *Config) bool {
 		return true
 	}
 	// match: (MOVDload [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem)
-	// cond: canMergeSym(sym1,sym2) 	&& ((off1+off2)%8==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym1) && !isAuto(sym1))
+	// cond: canMergeSym(sym1,sym2) 	&& is32Bit(off1+off2) && !isArg(mergeSym(sym1,sym2)) 	&& ((off1+off2)%8==0 || off1+off2<256 && off1+off2>-256 && !isAuto(mergeSym(sym1,sym2)))
 	// result: (MOVDload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
 	for {
 		off1 := v.AuxInt
@@ -4021,7 +4033,7 @@ func rewriteValueARM64_OpARM64MOVDload(v *Value, config *Config) bool {
 		sym2 := v_0.Aux
 		ptr := v_0.Args[0]
 		mem := v.Args[1]
-		if !(canMergeSym(sym1, sym2) && ((off1+off2)%8 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isArg(sym1) && !isAuto(sym1))) {
+		if !(canMergeSym(sym1, sym2) && is32Bit(off1+off2) && !isArg(mergeSym(sym1, sym2)) && ((off1+off2)%8 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isAuto(mergeSym(sym1, sym2)))) {
 			break
 		}
 		v.reset(OpARM64MOVDload)
@@ -4088,7 +4100,7 @@ func rewriteValueARM64_OpARM64MOVDstore(v *Value, config *Config) bool {
 	b := v.Block
 	_ = b
 	// match: (MOVDstore [off1] {sym} (ADDconst [off2] ptr) val mem)
-	// cond: (off1+off2)%8==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym) && !isAuto(sym)
+	// cond: is32Bit(off1+off2) && !isArg(sym) 	&& ((off1+off2)%8==0 || off1+off2<256 && off1+off2>-256 && !isAuto(sym))
 	// result: (MOVDstore [off1+off2] {sym} ptr val mem)
 	for {
 		off1 := v.AuxInt
@@ -4101,7 +4113,7 @@ func rewriteValueARM64_OpARM64MOVDstore(v *Value, config *Config) bool {
 		ptr := v_0.Args[0]
 		val := v.Args[1]
 		mem := v.Args[2]
-		if !((off1+off2)%8 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isArg(sym) && !isAuto(sym)) {
+		if !(is32Bit(off1+off2) && !isArg(sym) && ((off1+off2)%8 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isAuto(sym))) {
 			break
 		}
 		v.reset(OpARM64MOVDstore)
@@ -4113,7 +4125,7 @@ func rewriteValueARM64_OpARM64MOVDstore(v *Value, config *Config) bool {
 		return true
 	}
 	// match: (MOVDstore [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) val mem)
-	// cond: canMergeSym(sym1,sym2) 	&& ((off1+off2)%8==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym1) && !isAuto(sym1))
+	// cond: canMergeSym(sym1,sym2) 	&& is32Bit(off1+off2) && !isArg(mergeSym(sym1,sym2)) 	&& ((off1+off2)%8==0 || off1+off2<256 && off1+off2>-256 && !isAuto(mergeSym(sym1,sym2)))
 	// result: (MOVDstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem)
 	for {
 		off1 := v.AuxInt
@@ -4127,7 +4139,7 @@ func rewriteValueARM64_OpARM64MOVDstore(v *Value, config *Config) bool {
 		ptr := v_0.Args[0]
 		val := v.Args[1]
 		mem := v.Args[2]
-		if !(canMergeSym(sym1, sym2) && ((off1+off2)%8 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isArg(sym1) && !isAuto(sym1))) {
+		if !(canMergeSym(sym1, sym2) && is32Bit(off1+off2) && !isArg(mergeSym(sym1, sym2)) && ((off1+off2)%8 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isAuto(mergeSym(sym1, sym2)))) {
 			break
 		}
 		v.reset(OpARM64MOVDstore)
@@ -4166,7 +4178,7 @@ func rewriteValueARM64_OpARM64MOVDstorezero(v *Value, config *Config) bool {
 	b := v.Block
 	_ = b
 	// match: (MOVDstorezero [off1] {sym} (ADDconst [off2] ptr) mem)
-	// cond: (off1+off2)%2==8 || off1+off2<256 && off1+off2>-256 && !isArg(sym) && !isAuto(sym)
+	// cond: is32Bit(off1+off2) && !isArg(sym) 	&& ((off1+off2)%2==8 || off1+off2<256 && off1+off2>-256 && !isAuto(sym))
 	// result: (MOVDstorezero [off1+off2] {sym} ptr mem)
 	for {
 		off1 := v.AuxInt
@@ -4178,7 +4190,7 @@ func rewriteValueARM64_OpARM64MOVDstorezero(v *Value, config *Config) bool {
 		off2 := v_0.AuxInt
 		ptr := v_0.Args[0]
 		mem := v.Args[1]
-		if !((off1+off2)%2 == 8 || off1+off2 < 256 && off1+off2 > -256 && !isArg(sym) && !isAuto(sym)) {
+		if !(is32Bit(off1+off2) && !isArg(sym) && ((off1+off2)%2 == 8 || off1+off2 < 256 && off1+off2 > -256 && !isAuto(sym))) {
 			break
 		}
 		v.reset(OpARM64MOVDstorezero)
@@ -4189,7 +4201,7 @@ func rewriteValueARM64_OpARM64MOVDstorezero(v *Value, config *Config) bool {
 		return true
 	}
 	// match: (MOVDstorezero [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem)
-	// cond: canMergeSym(sym1,sym2) 	&& ((off1+off2)%8==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym1) && !isAuto(sym1))
+	// cond: canMergeSym(sym1,sym2) 	&& is32Bit(off1+off2) && !isArg(mergeSym(sym1,sym2)) 	&& ((off1+off2)%8==0 || off1+off2<256 && off1+off2>-256 && !isAuto(mergeSym(sym1,sym2)))
 	// result: (MOVDstorezero [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
 	for {
 		off1 := v.AuxInt
@@ -4202,7 +4214,7 @@ func rewriteValueARM64_OpARM64MOVDstorezero(v *Value, config *Config) bool {
 		sym2 := v_0.Aux
 		ptr := v_0.Args[0]
 		mem := v.Args[1]
-		if !(canMergeSym(sym1, sym2) && ((off1+off2)%8 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isArg(sym1) && !isAuto(sym1))) {
+		if !(canMergeSym(sym1, sym2) && is32Bit(off1+off2) && !isArg(mergeSym(sym1, sym2)) && ((off1+off2)%8 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isAuto(mergeSym(sym1, sym2)))) {
 			break
 		}
 		v.reset(OpARM64MOVDstorezero)
@@ -4218,7 +4230,7 @@ func rewriteValueARM64_OpARM64MOVHUload(v *Value, config *Config) bool {
 	b := v.Block
 	_ = b
 	// match: (MOVHUload [off1] {sym} (ADDconst [off2] ptr) mem)
-	// cond: (off1+off2)%2==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym) && !isAuto(sym)
+	// cond: is32Bit(off1+off2) && !isArg(sym) 	&& ((off1+off2)%2==0 || off1+off2<256 && off1+off2>-256 && !isAuto(sym))
 	// result: (MOVHUload [off1+off2] {sym} ptr mem)
 	for {
 		off1 := v.AuxInt
@@ -4230,7 +4242,7 @@ func rewriteValueARM64_OpARM64MOVHUload(v *Value, config *Config) bool {
 		off2 := v_0.AuxInt
 		ptr := v_0.Args[0]
 		mem := v.Args[1]
-		if !((off1+off2)%2 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isArg(sym) && !isAuto(sym)) {
+		if !(is32Bit(off1+off2) && !isArg(sym) && ((off1+off2)%2 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isAuto(sym))) {
 			break
 		}
 		v.reset(OpARM64MOVHUload)
@@ -4241,7 +4253,7 @@ func rewriteValueARM64_OpARM64MOVHUload(v *Value, config *Config) bool {
 		return true
 	}
 	// match: (MOVHUload [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem)
-	// cond: canMergeSym(sym1,sym2) 	&& ((off1+off2)%2==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym1) && !isAuto(sym1))
+	// cond: canMergeSym(sym1,sym2) 	&& is32Bit(off1+off2) && !isArg(mergeSym(sym1,sym2)) 	&& ((off1+off2)%2==0 || off1+off2<256 && off1+off2>-256 && !isAuto(mergeSym(sym1,sym2)))
 	// result: (MOVHUload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
 	for {
 		off1 := v.AuxInt
@@ -4254,7 +4266,7 @@ func rewriteValueARM64_OpARM64MOVHUload(v *Value, config *Config) bool {
 		sym2 := v_0.Aux
 		ptr := v_0.Args[0]
 		mem := v.Args[1]
-		if !(canMergeSym(sym1, sym2) && ((off1+off2)%2 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isArg(sym1) && !isAuto(sym1))) {
+		if !(canMergeSym(sym1, sym2) && is32Bit(off1+off2) && !isArg(mergeSym(sym1, sym2)) && ((off1+off2)%2 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isAuto(mergeSym(sym1, sym2)))) {
 			break
 		}
 		v.reset(OpARM64MOVHUload)
@@ -4357,7 +4369,7 @@ func rewriteValueARM64_OpARM64MOVHload(v *Value, config *Config) bool {
 	b := v.Block
 	_ = b
 	// match: (MOVHload [off1] {sym} (ADDconst [off2] ptr) mem)
-	// cond: (off1+off2)%2==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym) && !isAuto(sym)
+	// cond: is32Bit(off1+off2) && !isArg(sym) 	&& ((off1+off2)%2==0 || off1+off2<256 && off1+off2>-256 && !isAuto(sym))
 	// result: (MOVHload [off1+off2] {sym} ptr mem)
 	for {
 		off1 := v.AuxInt
@@ -4369,7 +4381,7 @@ func rewriteValueARM64_OpARM64MOVHload(v *Value, config *Config) bool {
 		off2 := v_0.AuxInt
 		ptr := v_0.Args[0]
 		mem := v.Args[1]
-		if !((off1+off2)%2 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isArg(sym) && !isAuto(sym)) {
+		if !(is32Bit(off1+off2) && !isArg(sym) && ((off1+off2)%2 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isAuto(sym))) {
 			break
 		}
 		v.reset(OpARM64MOVHload)
@@ -4380,7 +4392,7 @@ func rewriteValueARM64_OpARM64MOVHload(v *Value, config *Config) bool {
 		return true
 	}
 	// match: (MOVHload [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem)
-	// cond: canMergeSym(sym1,sym2) 	&& ((off1+off2)%2==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym1) && !isAuto(sym1))
+	// cond: canMergeSym(sym1,sym2) 	&& is32Bit(off1+off2) && !isArg(mergeSym(sym1,sym2)) 	&& ((off1+off2)%2==0 || off1+off2<256 && off1+off2>-256 && !isAuto(mergeSym(sym1,sym2)))
 	// result: (MOVHload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
 	for {
 		off1 := v.AuxInt
@@ -4393,7 +4405,7 @@ func rewriteValueARM64_OpARM64MOVHload(v *Value, config *Config) bool {
 		sym2 := v_0.Aux
 		ptr := v_0.Args[0]
 		mem := v.Args[1]
-		if !(canMergeSym(sym1, sym2) && ((off1+off2)%2 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isArg(sym1) && !isAuto(sym1))) {
+		if !(canMergeSym(sym1, sym2) && is32Bit(off1+off2) && !isArg(mergeSym(sym1, sym2)) && ((off1+off2)%2 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isAuto(mergeSym(sym1, sym2)))) {
 			break
 		}
 		v.reset(OpARM64MOVHload)
@@ -4520,7 +4532,7 @@ func rewriteValueARM64_OpARM64MOVHstore(v *Value, config *Config) bool {
 	b := v.Block
 	_ = b
 	// match: (MOVHstore [off1] {sym} (ADDconst [off2] ptr) val mem)
-	// cond: (off1+off2)%2==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym) && !isAuto(sym)
+	// cond: is32Bit(off1+off2) && !isArg(sym) 	&& ((off1+off2)%2==0 || off1+off2<256 && off1+off2>-256 && !isAuto(sym))
 	// result: (MOVHstore [off1+off2] {sym} ptr val mem)
 	for {
 		off1 := v.AuxInt
@@ -4533,7 +4545,7 @@ func rewriteValueARM64_OpARM64MOVHstore(v *Value, config *Config) bool {
 		ptr := v_0.Args[0]
 		val := v.Args[1]
 		mem := v.Args[2]
-		if !((off1+off2)%2 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isArg(sym) && !isAuto(sym)) {
+		if !(is32Bit(off1+off2) && !isArg(sym) && ((off1+off2)%2 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isAuto(sym))) {
 			break
 		}
 		v.reset(OpARM64MOVHstore)
@@ -4545,7 +4557,7 @@ func rewriteValueARM64_OpARM64MOVHstore(v *Value, config *Config) bool {
 		return true
 	}
 	// match: (MOVHstore [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) val mem)
-	// cond: canMergeSym(sym1,sym2) 	&& ((off1+off2)%2==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym1) && !isAuto(sym1))
+	// cond: canMergeSym(sym1,sym2) 	&& is32Bit(off1+off2) && !isArg(mergeSym(sym1,sym2)) 	&& ((off1+off2)%2==0 || off1+off2<256 && off1+off2>-256 && !isAuto(mergeSym(sym1,sym2)))
 	// result: (MOVHstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem)
 	for {
 		off1 := v.AuxInt
@@ -4559,7 +4571,7 @@ func rewriteValueARM64_OpARM64MOVHstore(v *Value, config *Config) bool {
 		ptr := v_0.Args[0]
 		val := v.Args[1]
 		mem := v.Args[2]
-		if !(canMergeSym(sym1, sym2) && ((off1+off2)%2 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isArg(sym1) && !isAuto(sym1))) {
+		if !(canMergeSym(sym1, sym2) && is32Bit(off1+off2) && !isArg(mergeSym(sym1, sym2)) && ((off1+off2)%2 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isAuto(mergeSym(sym1, sym2)))) {
 			break
 		}
 		v.reset(OpARM64MOVHstore)
@@ -4682,7 +4694,7 @@ func rewriteValueARM64_OpARM64MOVHstorezero(v *Value, config *Config) bool {
 	b := v.Block
 	_ = b
 	// match: (MOVHstorezero [off1] {sym} (ADDconst [off2] ptr) mem)
-	// cond: (off1+off2)%2==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym) && !isAuto(sym)
+	// cond: is32Bit(off1+off2) && !isArg(sym) 	&& ((off1+off2)%2==0 || off1+off2<256 && off1+off2>-256 && !isAuto(sym))
 	// result: (MOVHstorezero [off1+off2] {sym} ptr mem)
 	for {
 		off1 := v.AuxInt
@@ -4694,7 +4706,7 @@ func rewriteValueARM64_OpARM64MOVHstorezero(v *Value, config *Config) bool {
 		off2 := v_0.AuxInt
 		ptr := v_0.Args[0]
 		mem := v.Args[1]
-		if !((off1+off2)%2 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isArg(sym) && !isAuto(sym)) {
+		if !(is32Bit(off1+off2) && !isArg(sym) && ((off1+off2)%2 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isAuto(sym))) {
 			break
 		}
 		v.reset(OpARM64MOVHstorezero)
@@ -4705,7 +4717,7 @@ func rewriteValueARM64_OpARM64MOVHstorezero(v *Value, config *Config) bool {
 		return true
 	}
 	// match: (MOVHstorezero [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem)
-	// cond: canMergeSym(sym1,sym2) 	&& ((off1+off2)%2==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym1) && !isAuto(sym1))
+	// cond: canMergeSym(sym1,sym2) 	&& is32Bit(off1+off2) && !isArg(mergeSym(sym1,sym2)) 	&& ((off1+off2)%2==0 || off1+off2<256 && off1+off2>-256 && !isAuto(mergeSym(sym1,sym2)))
 	// result: (MOVHstorezero [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
 	for {
 		off1 := v.AuxInt
@@ -4718,7 +4730,7 @@ func rewriteValueARM64_OpARM64MOVHstorezero(v *Value, config *Config) bool {
 		sym2 := v_0.Aux
 		ptr := v_0.Args[0]
 		mem := v.Args[1]
-		if !(canMergeSym(sym1, sym2) && ((off1+off2)%2 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isArg(sym1) && !isAuto(sym1))) {
+		if !(canMergeSym(sym1, sym2) && is32Bit(off1+off2) && !isArg(mergeSym(sym1, sym2)) && ((off1+off2)%2 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isAuto(mergeSym(sym1, sym2)))) {
 			break
 		}
 		v.reset(OpARM64MOVHstorezero)
@@ -4734,7 +4746,7 @@ func rewriteValueARM64_OpARM64MOVWUload(v *Value, config *Config) bool {
 	b := v.Block
 	_ = b
 	// match: (MOVWUload [off1] {sym} (ADDconst [off2] ptr) mem)
-	// cond: (off1+off2)%4==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym) && !isAuto(sym)
+	// cond: is32Bit(off1+off2) && !isArg(sym) 	&& ((off1+off2)%4==0 || off1+off2<256 && off1+off2>-256 && !isAuto(sym))
 	// result: (MOVWUload [off1+off2] {sym} ptr mem)
 	for {
 		off1 := v.AuxInt
@@ -4746,7 +4758,7 @@ func rewriteValueARM64_OpARM64MOVWUload(v *Value, config *Config) bool {
 		off2 := v_0.AuxInt
 		ptr := v_0.Args[0]
 		mem := v.Args[1]
-		if !((off1+off2)%4 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isArg(sym) && !isAuto(sym)) {
+		if !(is32Bit(off1+off2) && !isArg(sym) && ((off1+off2)%4 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isAuto(sym))) {
 			break
 		}
 		v.reset(OpARM64MOVWUload)
@@ -4757,7 +4769,7 @@ func rewriteValueARM64_OpARM64MOVWUload(v *Value, config *Config) bool {
 		return true
 	}
 	// match: (MOVWUload [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem)
-	// cond: canMergeSym(sym1,sym2) 	&& ((off1+off2)%4==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym1) && !isAuto(sym1))
+	// cond: canMergeSym(sym1,sym2) 	&& is32Bit(off1+off2) && !isArg(mergeSym(sym1,sym2)) 	&& ((off1+off2)%4==0 || off1+off2<256 && off1+off2>-256 && !isAuto(mergeSym(sym1,sym2)))
 	// result: (MOVWUload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
 	for {
 		off1 := v.AuxInt
@@ -4770,7 +4782,7 @@ func rewriteValueARM64_OpARM64MOVWUload(v *Value, config *Config) bool {
 		sym2 := v_0.Aux
 		ptr := v_0.Args[0]
 		mem := v.Args[1]
-		if !(canMergeSym(sym1, sym2) && ((off1+off2)%4 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isArg(sym1) && !isAuto(sym1))) {
+		if !(canMergeSym(sym1, sym2) && is32Bit(off1+off2) && !isArg(mergeSym(sym1, sym2)) && ((off1+off2)%4 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isAuto(mergeSym(sym1, sym2)))) {
 			break
 		}
 		v.reset(OpARM64MOVWUload)
@@ -4897,7 +4909,7 @@ func rewriteValueARM64_OpARM64MOVWload(v *Value, config *Config) bool {
 	b := v.Block
 	_ = b
 	// match: (MOVWload [off1] {sym} (ADDconst [off2] ptr) mem)
-	// cond: (off1+off2)%4==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym) && !isAuto(sym)
+	// cond: is32Bit(off1+off2) && !isArg(sym) 	&& ((off1+off2)%4==0 || off1+off2<256 && off1+off2>-256 && !isAuto(sym))
 	// result: (MOVWload [off1+off2] {sym} ptr mem)
 	for {
 		off1 := v.AuxInt
@@ -4909,7 +4921,7 @@ func rewriteValueARM64_OpARM64MOVWload(v *Value, config *Config) bool {
 		off2 := v_0.AuxInt
 		ptr := v_0.Args[0]
 		mem := v.Args[1]
-		if !((off1+off2)%4 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isArg(sym) && !isAuto(sym)) {
+		if !(is32Bit(off1+off2) && !isArg(sym) && ((off1+off2)%4 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isAuto(sym))) {
 			break
 		}
 		v.reset(OpARM64MOVWload)
@@ -4920,7 +4932,7 @@ func rewriteValueARM64_OpARM64MOVWload(v *Value, config *Config) bool {
 		return true
 	}
 	// match: (MOVWload [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem)
-	// cond: canMergeSym(sym1,sym2) 	&& ((off1+off2)%4==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym1) && !isAuto(sym1))
+	// cond: canMergeSym(sym1,sym2) 	&& is32Bit(off1+off2) && !isArg(mergeSym(sym1,sym2)) 	&& ((off1+off2)%4==0 || off1+off2<256 && off1+off2>-256 && !isAuto(mergeSym(sym1,sym2)))
 	// result: (MOVWload [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
 	for {
 		off1 := v.AuxInt
@@ -4933,7 +4945,7 @@ func rewriteValueARM64_OpARM64MOVWload(v *Value, config *Config) bool {
 		sym2 := v_0.Aux
 		ptr := v_0.Args[0]
 		mem := v.Args[1]
-		if !(canMergeSym(sym1, sym2) && ((off1+off2)%4 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isArg(sym1) && !isAuto(sym1))) {
+		if !(canMergeSym(sym1, sym2) && is32Bit(off1+off2) && !isArg(mergeSym(sym1, sym2)) && ((off1+off2)%4 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isAuto(mergeSym(sym1, sym2)))) {
 			break
 		}
 		v.reset(OpARM64MOVWload)
@@ -5108,7 +5120,7 @@ func rewriteValueARM64_OpARM64MOVWstore(v *Value, config *Config) bool {
 	b := v.Block
 	_ = b
 	// match: (MOVWstore [off1] {sym} (ADDconst [off2] ptr) val mem)
-	// cond: (off1+off2)%4==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym) && !isAuto(sym)
+	// cond: is32Bit(off1+off2) && !isArg(sym) 	&& ((off1+off2)%4==0 || off1+off2<256 && off1+off2>-256 && !isAuto(sym))
 	// result: (MOVWstore [off1+off2] {sym} ptr val mem)
 	for {
 		off1 := v.AuxInt
@@ -5121,7 +5133,7 @@ func rewriteValueARM64_OpARM64MOVWstore(v *Value, config *Config) bool {
 		ptr := v_0.Args[0]
 		val := v.Args[1]
 		mem := v.Args[2]
-		if !((off1+off2)%4 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isArg(sym) && !isAuto(sym)) {
+		if !(is32Bit(off1+off2) && !isArg(sym) && ((off1+off2)%4 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isAuto(sym))) {
 			break
 		}
 		v.reset(OpARM64MOVWstore)
@@ -5133,7 +5145,7 @@ func rewriteValueARM64_OpARM64MOVWstore(v *Value, config *Config) bool {
 		return true
 	}
 	// match: (MOVWstore [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) val mem)
-	// cond: canMergeSym(sym1,sym2) 	&& ((off1+off2)%4==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym1) && !isAuto(sym1))
+	// cond: canMergeSym(sym1,sym2) 	&& is32Bit(off1+off2) && !isArg(mergeSym(sym1,sym2)) 	&& ((off1+off2)%4==0 || off1+off2<256 && off1+off2>-256 && !isAuto(mergeSym(sym1,sym2)))
 	// result: (MOVWstore [off1+off2] {mergeSym(sym1,sym2)} ptr val mem)
 	for {
 		off1 := v.AuxInt
@@ -5147,7 +5159,7 @@ func rewriteValueARM64_OpARM64MOVWstore(v *Value, config *Config) bool {
 		ptr := v_0.Args[0]
 		val := v.Args[1]
 		mem := v.Args[2]
-		if !(canMergeSym(sym1, sym2) && ((off1+off2)%4 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isArg(sym1) && !isAuto(sym1))) {
+		if !(canMergeSym(sym1, sym2) && is32Bit(off1+off2) && !isArg(mergeSym(sym1, sym2)) && ((off1+off2)%4 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isAuto(mergeSym(sym1, sym2)))) {
 			break
 		}
 		v.reset(OpARM64MOVWstore)
@@ -5228,7 +5240,7 @@ func rewriteValueARM64_OpARM64MOVWstorezero(v *Value, config *Config) bool {
 	b := v.Block
 	_ = b
 	// match: (MOVWstorezero [off1] {sym} (ADDconst [off2] ptr) mem)
-	// cond: (off1+off2)%4==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym) && !isAuto(sym)
+	// cond: is32Bit(off1+off2) && !isArg(sym) 	&& ((off1+off2)%4==0 || off1+off2<256 && off1+off2>-256 && !isAuto(sym))
 	// result: (MOVWstorezero [off1+off2] {sym} ptr mem)
 	for {
 		off1 := v.AuxInt
@@ -5240,7 +5252,7 @@ func rewriteValueARM64_OpARM64MOVWstorezero(v *Value, config *Config) bool {
 		off2 := v_0.AuxInt
 		ptr := v_0.Args[0]
 		mem := v.Args[1]
-		if !((off1+off2)%4 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isArg(sym) && !isAuto(sym)) {
+		if !(is32Bit(off1+off2) && !isArg(sym) && ((off1+off2)%4 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isAuto(sym))) {
 			break
 		}
 		v.reset(OpARM64MOVWstorezero)
@@ -5251,7 +5263,7 @@ func rewriteValueARM64_OpARM64MOVWstorezero(v *Value, config *Config) bool {
 		return true
 	}
 	// match: (MOVWstorezero [off1] {sym1} (MOVDaddr [off2] {sym2} ptr) mem)
-	// cond: canMergeSym(sym1,sym2) 	&& ((off1+off2)%4==0 || off1+off2<256 && off1+off2>-256 && !isArg(sym1) && !isAuto(sym1))
+	// cond: canMergeSym(sym1,sym2) 	&& is32Bit(off1+off2) && !isArg(mergeSym(sym1,sym2)) 	&& ((off1+off2)%4==0 || off1+off2<256 && off1+off2>-256 && !isAuto(mergeSym(sym1,sym2)))
 	// result: (MOVWstorezero [off1+off2] {mergeSym(sym1,sym2)} ptr mem)
 	for {
 		off1 := v.AuxInt
@@ -5264,7 +5276,7 @@ func rewriteValueARM64_OpARM64MOVWstorezero(v *Value, config *Config) bool {
 		sym2 := v_0.Aux
 		ptr := v_0.Args[0]
 		mem := v.Args[1]
-		if !(canMergeSym(sym1, sym2) && ((off1+off2)%4 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isArg(sym1) && !isAuto(sym1))) {
+		if !(canMergeSym(sym1, sym2) && is32Bit(off1+off2) && !isArg(mergeSym(sym1, sym2)) && ((off1+off2)%4 == 0 || off1+off2 < 256 && off1+off2 > -256 && !isAuto(mergeSym(sym1, sym2)))) {
 			break
 		}
 		v.reset(OpARM64MOVWstorezero)

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

@@ -712,7 +712,7 @@ func rewriteValueMIPS_OpAtomicAnd8(v *Value, config *Config) bool {
 	_ = b
 	// match: (AtomicAnd8  ptr val mem)
 	// cond: !config.BigEndian
-	// result: (LoweredAtomicAnd (AND <config.fe.TypeUInt32().PtrTo()> (MOVWconst [^3]) ptr) 		(OR <config.fe.TypeUInt32()> (SLL <config.fe.TypeUInt32()> (ZeroExt8to32 val) 			(SLLconst <config.fe.TypeUInt32()> [3] 				(ANDconst  <config.fe.TypeUInt32()> [3] ptr))) 		(NORconst [0] <config.fe.TypeUInt32()> (SLL <config.fe.TypeUInt32()> 			(MOVWconst [0xff]) (SLLconst <config.fe.TypeUInt32()> [3] 				(ANDconst <config.fe.TypeUInt32()> [3] 					(XORconst <config.fe.TypeUInt32()> [3] ptr)))))) mem)
+	// result: (LoweredAtomicAnd (AND <config.fe.TypeUInt32().PtrTo()> (MOVWconst [^3]) ptr) 		(OR <config.fe.TypeUInt32()> (SLL <config.fe.TypeUInt32()> (ZeroExt8to32 val) 			(SLLconst <config.fe.TypeUInt32()> [3] 				(ANDconst  <config.fe.TypeUInt32()> [3] ptr))) 		(NORconst [0] <config.fe.TypeUInt32()> (SLL <config.fe.TypeUInt32()> 			(MOVWconst [0xff]) (SLLconst <config.fe.TypeUInt32()> [3] 				(ANDconst <config.fe.TypeUInt32()> [3] ptr))))) mem)
 	for {
 		ptr := v.Args[0]
 		val := v.Args[1]
@@ -750,10 +750,7 @@ func rewriteValueMIPS_OpAtomicAnd8(v *Value, config *Config) bool {
 		v10.AuxInt = 3
 		v11 := b.NewValue0(v.Line, OpMIPSANDconst, config.fe.TypeUInt32())
 		v11.AuxInt = 3
-		v12 := b.NewValue0(v.Line, OpMIPSXORconst, config.fe.TypeUInt32())
-		v12.AuxInt = 3
-		v12.AddArg(ptr)
-		v11.AddArg(v12)
+		v11.AddArg(ptr)
 		v10.AddArg(v11)
 		v8.AddArg(v10)
 		v7.AddArg(v8)

src/cmd/compile/internal/ssa/rewritePPC64.go

@@ -1543,12 +1543,12 @@ func rewriteValuePPC64_OpGeq16U(v *Value, config *Config) bool {
 	_ = b
 	// match: (Geq16U x y)
 	// cond:
-	// result: (GreaterEqual (CMPU (ZeroExt16to32 x) (ZeroExt16to32 y)))
+	// result: (GreaterEqual (CMPWU (ZeroExt16to32 x) (ZeroExt16to32 y)))
 	for {
 		x := v.Args[0]
 		y := v.Args[1]
 		v.reset(OpPPC64GreaterEqual)
-		v0 := b.NewValue0(v.Line, OpPPC64CMPU, TypeFlags)
+		v0 := b.NewValue0(v.Line, OpPPC64CMPWU, TypeFlags)
 		v1 := b.NewValue0(v.Line, OpZeroExt16to32, config.fe.TypeUInt32())
 		v1.AddArg(x)
 		v0.AddArg(v1)
@@ -1598,12 +1598,12 @@ func rewriteValuePPC64_OpGeq32U(v *Value, config *Config) bool {
 	_ = b
 	// match: (Geq32U x y)
 	// cond:
-	// result: (GreaterEqual (CMPU x y))
+	// result: (GreaterEqual (CMPWU x y))
 	for {
 		x := v.Args[0]
 		y := v.Args[1]
 		v.reset(OpPPC64GreaterEqual)
-		v0 := b.NewValue0(v.Line, OpPPC64CMPU, TypeFlags)
+		v0 := b.NewValue0(v.Line, OpPPC64CMPWU, TypeFlags)
 		v0.AddArg(x)
 		v0.AddArg(y)
 		v.AddArg(v0)
@@ -1687,12 +1687,12 @@ func rewriteValuePPC64_OpGeq8U(v *Value, config *Config) bool {
 	_ = b
 	// match: (Geq8U x y)
 	// cond:
-	// result: (GreaterEqual (CMPU (ZeroExt8to32 x) (ZeroExt8to32 y)))
+	// result: (GreaterEqual (CMPWU (ZeroExt8to32 x) (ZeroExt8to32 y)))
 	for {
 		x := v.Args[0]
 		y := v.Args[1]
 		v.reset(OpPPC64GreaterEqual)
-		v0 := b.NewValue0(v.Line, OpPPC64CMPU, TypeFlags)
+		v0 := b.NewValue0(v.Line, OpPPC64CMPWU, TypeFlags)
 		v1 := b.NewValue0(v.Line, OpZeroExt8to32, config.fe.TypeUInt32())
 		v1.AddArg(x)
 		v0.AddArg(v1)

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

@@ -14,6 +14,7 @@ const (
 	ScoreMemory
 	ScoreDefault
 	ScoreFlags
+	ScoreSelectCall
 	ScoreControl // towards bottom of block
 )
 
@@ -110,10 +111,25 @@ func schedule(f *Func) {
 				// We want all the vardefs next.
 				score[v.ID] = ScoreVarDef
 			case v.Type.IsMemory():
-				// Schedule stores as early as possible. This tends to
-				// reduce register pressure. It also helps make sure
-				// VARDEF ops are scheduled before the corresponding LEA.
-				score[v.ID] = ScoreMemory
+				// Don't schedule independent operations after call to those functions.
+				// runtime.selectgo will jump to next instruction after this call,
+				// causing extra execution of those operations. Prevent it, by setting
+				// priority to high value.
+				if (v.Op == OpAMD64CALLstatic || v.Op == OpPPC64CALLstatic ||
+					v.Op == OpARMCALLstatic || v.Op == OpARM64CALLstatic ||
+					v.Op == Op386CALLstatic || v.Op == OpMIPS64CALLstatic ||
+					v.Op == OpS390XCALLstatic || v.Op == OpMIPSCALLstatic) &&
+					(isSameSym(v.Aux, "runtime.selectrecv") ||
+						isSameSym(v.Aux, "runtime.selectrecv2") ||
+						isSameSym(v.Aux, "runtime.selectsend") ||
+						isSameSym(v.Aux, "runtime.selectdefault")) {
+					score[v.ID] = ScoreSelectCall
+				} else {
+					// Schedule stores as early as possible. This tends to
+					// reduce register pressure. It also helps make sure
+					// VARDEF ops are scheduled before the corresponding LEA.
+					score[v.ID] = ScoreMemory
+				}
 			case v.Op == OpSelect0 || v.Op == OpSelect1:
 				// Schedule the pseudo-op of reading part of a tuple
 				// immediately after the tuple-generating op, since
@@ -132,19 +148,20 @@ func schedule(f *Func) {
 		}
 	}
 
+	// TODO: make this logic permanent in types.IsMemory?
+	isMem := func(v *Value) bool {
+		return v.Type.IsMemory() || v.Type.IsTuple() && v.Type.FieldType(1).IsMemory()
+	}
+
 	for _, b := range f.Blocks {
 		// Find store chain for block.
 		// Store chains for different blocks overwrite each other, so
 		// the calculated store chain is good only for this block.
 		for _, v := range b.Values {
-			if v.Op != OpPhi && v.Type.IsMemory() {
-				mem := v
-				if v.Op == OpSelect1 {
-					v = v.Args[0]
-				}
+			if v.Op != OpPhi && isMem(v) {
 				for _, w := range v.Args {
-					if w.Type.IsMemory() {
-						nextMem[w.ID] = mem
+					if isMem(w) {
+						nextMem[w.ID] = v
 					}
 				}
 			}
@@ -163,15 +180,15 @@ func schedule(f *Func) {
 					uses[w.ID]++
 				}
 				// Any load must come before the following store.
-				if v.Type.IsMemory() || !w.Type.IsMemory() {
-					continue // not a load
+				if !isMem(v) && isMem(w) {
+					// v is a load.
+					s := nextMem[w.ID]
+					if s == nil || s.Block != b {
+						continue
+					}
+					additionalArgs[s.ID] = append(additionalArgs[s.ID], v)
+					uses[v.ID]++
 				}
-				s := nextMem[w.ID]
-				if s == nil || s.Block != b {
-					continue
-				}
-				additionalArgs[s.ID] = append(additionalArgs[s.ID], v)
-				uses[v.ID]++
 			}
 		}