Extends avo to support most AVX-512 instruction sets.
The instruction type is extended to support suffixes. The K family of opmask
registers is added to the register package, and the operand package is updated
to support the new operand types. Move instruction deduction in `Load` and
`Store` is extended to support KMOV* and VMOV* forms.
Internal code generation packages were overhauled. Instruction database loading
required various messy changes to account for the additional complexities of the
AVX-512 instruction sets. The internal/api package was added to introduce a
separation between instruction forms in the database, and the functions avo
provides to create them. This was required since with instruction suffixes there
is no longer a one-to-one mapping between instruction constructors and opcodes.
AVX-512 bloated generated source code size substantially, initially increasing
compilation and CI test times to an unacceptable level. Two changes were made to
address this:
1. Instruction constructors in the `x86` package moved to an optab-based
approach. This compiles substantially faster than the verbose code
generation we had before.
2. The most verbose code-generated tests are moved under build tags and
limited to a stress test mode. Stress test builds are run on
schedule but not in regular CI.
An example of AVX-512 accelerated 16-lane MD5 is provided to demonstrate and
test the new functionality.
Updates #20#163#229
Co-authored-by: Vaughn Iverson <vsivsi@yahoo.com>
Currently `avo` uses `BP` as a standard general-purpose register. However, `BP` is used for the frame pointer and should be callee-save. Under some circumstances, the Go assembler will do this automatically, but not always. At the moment `avo` can produce code that clobbers the `BP` register. Since Go 1.16 this code will also fail a new `go vet` check.
This PR provides a (currently sub-optimal) fix for the issue. It introduces an `EnsureBasePointerCalleeSaved` pass which will check if the base pointer is written to by a function, and if so will artificially ensure that the function has a non-zero frame size. This will trigger the Go assembler to automatically save and restore the BP register.
In addition, we update the `asmdecl` tool to `asmvet`, which includes the `framepointer` vet check.
Updates #156
An update to gofumpt caused the lint job to start failing:
https://github.com/mmcloughlin/avo/runs/638734114?check_suite_focus=true#step:7:688
The underlying issue is that the "join parameters" rule added to gofumpt
disagrees with output from go/types. It's possible to fix, but I don't
think it's important enough to deal with right now. Instead I'll pin the
gofumpt dependency to a version before that rule was added.
Pinning tool dependencies is a good idea anyway.
Updates modules handling so build passes in Go 1.13. Bypasses modules for tools install in bootstrap. Upgrades golangci-lint version to avoid bug with v1.17.1 under Go 1.13.
Fixes#95
Run asmfmt suring linting and confirm git repository isn't dirty.
This introduces a developer tools dependency on asmfmt, but not a
runtime dependency.
Updates #8
Attempt to fix a stream of CI failures that started on the addition of
Go modules.
I wonder if we are falling foul of this warning in the wiki
https://github.com/golang/go/wiki/Modules.
"The module cache in Go 1.11 can sometimes cause various errors,
primarily if there were previously network issues or multiple go
commands executing in parallel (see #26794, which is addressed for Go
1.12). As a troubleshooting step, you can copy $GOPATH/pkg/mod to a
backup directory (in case further investigation is warranted later), run
go clean -modcache, and then see whether the original problem persists."
Updates #21
We depend upon data files from golang arch sub-repo and the Opcodes XML
database (behind PeachPy). This change includes their LICENSE files.
Updates #22
Adds methods for referencing sub- or super-registers. For example, for
general purpose registers you can now reference As8(), As16(), ... and
for vector AsX(), AsY(), AsZ().
Closes#1
