all: AVX-512 (#217)

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>

reg/collection.go

@@ -52,3 +52,6 @@ func (c *Collection) ZMM() VecVirtual { return c.Vec(S512) }
 
 // Vec allocates and returns a vector register of the given width.
 func (c *Collection) Vec(s Spec) VecVirtual { return newvecv(c.VirtualRegister(KindVector, s)) }
+
+// K allocates and returns an opmask register.
+func (c *Collection) K() OpmaskVirtual { return newopmaskv(c.VirtualRegister(KindOpmask, S64)) }

reg/reg_test.go

@@ -19,6 +19,7 @@ func TestIDIsVirtual(t *testing.T) {
 	cases := []Virtual{
 		GeneralPurpose.Virtual(42, S64),
 		Vector.Virtual(42, S128),
+		Opmask.Virtual(42, S64),
 	}
 	for _, r := range cases {
 		if !r.ID().IsVirtual() {
@@ -28,7 +29,7 @@ func TestIDIsVirtual(t *testing.T) {
 }
 
 func TestIDIsPhysical(t *testing.T) {
-	cases := []Physical{AL, AH, AX, EAX, RAX, X1, Y2, Z31}
+	cases := []Physical{AL, AH, AX, EAX, RAX, X1, Y2, Z31, K1}
 	for _, r := range cases {
 		if !r.ID().IsPhysical() {
 			t.FailNow()
@@ -98,6 +99,9 @@ func TestFamilyLookup(t *testing.T) {
 		{Vector, 27, S512, Z27},
 		{Vector, 1, S16, nil},
 		{Vector, 299, S256, nil},
+		{Opmask, 1, S64, K1},
+		{Opmask, 8, S64, nil},
+		{Opmask, 0, S64, K0},
 	}
 	for _, c := range cases {
 		got := c.Family.Lookup(c.ID, c.Spec)
@@ -156,6 +160,8 @@ func TestLookupPhysical(t *testing.T) {
 		{KindVector, 7, S128, X7},
 		{KindVector, 17, S256, Y17},
 		{KindVector, 27, S512, Z27},
+
+		{KindOpmask, 1, S64, K1},
 	}
 	for _, c := range cases {
 		if got := LookupPhysical(c.Kind, c.Index, c.Spec); !Equal(got, c.Expect) {
@@ -165,7 +171,7 @@ func TestLookupPhysical(t *testing.T) {
 }
 
 func TestLookupIDSelf(t *testing.T) {
-	cases := []Physical{AL, AH, AX, EAX, RAX, X1, Y2, Z31}
+	cases := []Physical{AL, AH, AX, EAX, RAX, X1, Y2, Z31, K1}
 	for _, r := range cases {
 		if got := LookupID(r.ID(), r.spec()); !Equal(got, r) {
 			t.FailNow()

reg/x86.go

@@ -5,6 +5,7 @@ const (
 	KindPseudo Kind = iota
 	KindGP
 	KindVector
+	KindOpmask
 )
 
 // Declare register families.
@@ -12,11 +13,13 @@ var (
 	Pseudo         = &Family{Kind: KindPseudo}
 	GeneralPurpose = &Family{Kind: KindGP}
 	Vector         = &Family{Kind: KindVector}
+	Opmask         = &Family{Kind: KindOpmask}
 
 	Families = []*Family{
 		Pseudo,
 		GeneralPurpose,
 		Vector,
+		Opmask,
 	}
 )
 
@@ -329,3 +332,52 @@ var (
 	Z30 = vec(S512, 30, "Z30")
 	Z31 = vec(S512, 31, "Z31")
 )
+
+// OpmaskPhysical is a opmask physical register.
+type OpmaskPhysical interface {
+	Physical
+}
+
+type opmaskp struct {
+	Physical
+}
+
+func newopmaskp(r Physical) OpmaskPhysical { return opmaskp{Physical: r} }
+
+// OpmaskVirtual is a virtual opmask register.
+type OpmaskVirtual interface {
+	Virtual
+}
+
+type opmaskv struct {
+	Virtual
+}
+
+func newopmaskv(v Virtual) OpmaskVirtual { return opmaskv{Virtual: v} }
+
+func opmask(s Spec, id Index, name string, flags ...Info) OpmaskPhysical {
+	r := newopmaskp(newregister(Opmask, s, id, name, flags...))
+	Opmask.add(r)
+	return r
+}
+
+// Opmask registers.
+//
+// Note that while K0 is a physical opmask register (it is a valid opmask source
+// and destination operand), it cannot be used as an opmask predicate value
+// because in that context K0 means "all true" or "no mask" regardless of the
+// actual contents of the physical register. For that reason, K0 should never be
+// assigned as a "general purpose" opmask register. However, it can be
+// explicitly operated upon by name as non-predicate operand, for example to
+// hold a constant or temporary value during calculations on other opmask
+// registers.
+var (
+	K0 = opmask(S64, 0, "K0", Restricted)
+	K1 = opmask(S64, 1, "K1")
+	K2 = opmask(S64, 2, "K2")
+	K3 = opmask(S64, 3, "K3")
+	K4 = opmask(S64, 4, "K4")
+	K5 = opmask(S64, 5, "K5")
+	K6 = opmask(S64, 6, "K6")
+	K7 = opmask(S64, 7, "K7")
+)