feat(arch): add the AVX512-FP16 scalar family to the amd64 extension layer

Assisted-by: GLM 5.3 Flash

asm/extension_amd64_test.go

@@ -12,8 +12,9 @@ import (
 )
 
 // TestAmd64ExtensionRegistry checks the mnemonic lookup for the amd64 layer:
-// one mnemonic across several vector lengths resolves to every entry, the
-// lookup is case-insensitive, and the counts match the registered families.
+// one mnemonic across several vector lengths or W bits resolves to every
+// entry, the lookup is case-insensitive, and the counts match the registered
+// families.
 func TestAmd64ExtensionRegistry(t *testing.T) {
 	for _, tt := range []struct {
 		mnem  string
@@ -24,6 +25,14 @@ func TestAmd64ExtensionRegistry(t *testing.T) {
 		{"VDPBF16PS", 3},
 		{"VP2INTERSECTD", 3},
 		{"VP2INTERSECTQ", 3},
+		{"VMOVSH", 1},
+		{"VMOVW", 2},
+		{"VADDSH", 1},
+		{"VSQRTSH", 1},
+		{"VCOMISH", 1},
+		{"VCVTSH2SS", 1},
+		{"VCVTSI2SH", 2},
+		{"VCVTSH2SI", 2},
 	} {
 		cands, ok := LookupExtension(arch.AMD64, tt.mnem)
 		if !ok {
@@ -37,8 +46,8 @@ func TestAmd64ExtensionRegistry(t *testing.T) {
 			t.Errorf("the %s lookup is not case-insensitive", tt.mnem)
 		}
 	}
-	if got := arch.Extensions(arch.AMD64); len(got) != 15 {
-		t.Errorf("the amd64 layer registers %d instructions, want 15", len(got))
+	if got := arch.Extensions(arch.AMD64); len(got) != 39 {
+		t.Errorf("the amd64 layer registers %d instructions, want 39", len(got))
 	}
 	if _, ok := LookupExtension(arch.AMD64, "NOSUCHINSTR"); ok {
 		t.Error("a non-extended mnemonic resolved")
@@ -87,6 +96,21 @@ func TestEncodeExtensionAmd64(t *testing.T) {
 		{"intersect into a mask", "VP2INTERSECTD",
 			[]arch.ExtOperand{arch.ExtYmm(2), arch.ExtYmm(1), arch.ExtMask(2)},
 			"62f26f2868d1"},
+		{"scalar fp16 add, high registers", "VADDSH",
+			[]arch.ExtOperand{arch.ExtXmm(29), arch.ExtXmm(28), arch.ExtXmm(30)},
+			"6205160058f4"},
+		{"scalar compare", "VCOMISH",
+			[]arch.ExtOperand{arch.ExtXmm(29), arch.ExtXmm(30)},
+			"62057c082ff5"},
+		{"integer into a scalar fp16", "VCVTSI2SH",
+			[]arch.ExtOperand{arch.ExtXmm(29), arch.ExtGpr64(12), arch.ExtXmm(30)},
+			"624596002af4"},
+		{"scalar fp16 into an integer", "VCVTSH2SI",
+			[]arch.ExtOperand{arch.ExtXmm(30), arch.ExtGpr32(2)},
+			"62957e082dd6"},
+		{"word move into an xmm", "VMOVW",
+			[]arch.ExtOperand{arch.ExtGpr64(12), arch.ExtXmm(30)},
+			"62457d086ef4"},
 	} {
 		got, err := EncodeExtension(arch.AMD64, tt.mnem, tt.ops...)
 		if err != nil {