cmd/asm: add support for the rest of loong64 unary bitops

All remaining unary bitop instructions in the LoongArch v1.00 base ISA
are added with this change.

While at it, add the missing W suffix to the current CLO/CLZ names. They
are not used anywhere as far as we know, so no breakage is expected.
Also, stop reusing SLL's instruction format for simplicity, in favor of
a new but trivial instruction format case.

This patch is a copy of CL 477717.
Co-authored-by: WANG Xuerui <git@xen0n.name>

Change-Id: Idbcaca25dda1ed313674ef8b26da722e8d7151c0
Reviewed-on: https://go-review.googlesource.com/c/go/+/623876
Reviewed-by: abner chenc <chenguoqi@loongson.cn>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Carlos Amedee <carlos@golang.org>
Reviewed-by: Cherry Mui <cherryyz@google.com>

3 files changed, 113 insertions, 11 deletions

diff --git a/src/cmd/internal/obj/loong64/a.out.go b/src/cmd/internal/obj/loong64/a.out.go
index a2a809cbc0..6653b7fe98 100644
--- a/src/cmd/internal/obj/loong64/a.out.go
+++ b/src/cmd/internal/obj/loong64/a.out.go
@@ -249,8 +249,6 @@ const (
 
 	ABNE
 	ABREAK
-	ACLO
-	ACLZ
 
 	ACMPEQD
 	ACMPEQF
@@ -439,6 +437,38 @@ const (
 	AAMMINDBWU
 	AAMMINDBVU
 
+	// 2.2.3.1
+	AEXTWB
+	AEXTWH
+
+	// 2.2.3.2
+	ACLOW
+	ACLOV
+	ACLZW
+	ACLZV
+	ACTOW
+	ACTOV
+	ACTZW
+	ACTZV
+
+	// 2.2.3.4
+	AREVBV
+	AREVB2W
+	AREVB4H
+	AREVB2H
+
+	// 2.2.3.5
+	AREVH2W
+	AREVHV
+
+	// 2.2.3.6
+	ABITREV4B
+	ABITREV8B
+
+	// 2.2.3.7
+	ABITREVW
+	ABITREVV
+
 	// 2.2.3.8
 	ABSTRINSW
 	ABSTRINSV
diff --git a/src/cmd/internal/obj/loong64/anames.go b/src/cmd/internal/obj/loong64/anames.go
index 2dc6319c4b..2cbf305196 100644
--- a/src/cmd/internal/obj/loong64/anames.go
+++ b/src/cmd/internal/obj/loong64/anames.go
@@ -22,8 +22,6 @@ var Anames = []string{
 	"BFPT",
 	"BNE",
 	"BREAK",
-	"CLO",
-	"CLZ",
 	"CMPEQD",
 	"CMPEQF",
 	"CMPGED",
@@ -177,6 +175,26 @@ var Anames = []string{
 	"AMMAXDBVU",
 	"AMMINDBWU",
 	"AMMINDBVU",
+	"EXTWB",
+	"EXTWH",
+	"CLOW",
+	"CLOV",
+	"CLZW",
+	"CLZV",
+	"CTOW",
+	"CTOV",
+	"CTZW",
+	"CTZV",
+	"REVBV",
+	"REVB2W",
+	"REVB4H",
+	"REVB2H",
+	"REVH2W",
+	"REVHV",
+	"BITREV4B",
+	"BITREV8B",
+	"BITREVW",
+	"BITREVV",
 	"BSTRINSW",
 	"BSTRINSV",
 	"BSTRPICKW",
diff --git a/src/cmd/internal/obj/loong64/asm.go b/src/cmd/internal/obj/loong64/asm.go
index 926b9952a2..985c054ece 100644
--- a/src/cmd/internal/obj/loong64/asm.go
+++ b/src/cmd/internal/obj/loong64/asm.go
@@ -86,7 +86,7 @@ var optab = []Optab{
 	{AADDF, C_FREG, C_FREG, C_NONE, C_FREG, C_NONE, 2, 4, 0, 0},
 	{ACMPEQF, C_FREG, C_FREG, C_NONE, C_FCCREG, C_NONE, 2, 4, 0, 0},
 
-	{ACLO, C_REG, C_NONE, C_NONE, C_REG, C_NONE, 9, 4, 0, 0},
+	{ACLOW, C_REG, C_NONE, C_NONE, C_REG, C_NONE, 9, 4, 0, 0},
 	{AABSF, C_FREG, C_NONE, C_NONE, C_FREG, C_NONE, 9, 4, 0, 0},
 	{AMOVVF, C_FREG, C_NONE, C_NONE, C_FREG, C_NONE, 9, 4, 0, 0},
 	{AMOVF, C_FREG, C_NONE, C_NONE, C_FREG, C_NONE, 9, 4, 0, 0},
@@ -1207,8 +1207,26 @@ func buildop(ctxt *obj.Link) {
 			opset(ARDTIMEHW, r0)
 			opset(ARDTIMED, r0)
 
-		case ACLO:
-			opset(ACLZ, r0)
+		case ACLOW:
+			opset(ACLZW, r0)
+			opset(ACTOW, r0)
+			opset(ACTZW, r0)
+			opset(ACLOV, r0)
+			opset(ACLZV, r0)
+			opset(ACTOV, r0)
+			opset(ACTZV, r0)
+			opset(AREVB2H, r0)
+			opset(AREVB4H, r0)
+			opset(AREVB2W, r0)
+			opset(AREVBV, r0)
+			opset(AREVH2W, r0)
+			opset(AREVHV, r0)
+			opset(ABITREV4B, r0)
+			opset(ABITREV8B, r0)
+			opset(ABITREVW, r0)
+			opset(ABITREVV, r0)
+			opset(AEXTWB, r0)
+			opset(AEXTWH, r0)
 			opset(ACPUCFG, r0)
 
 		case ATEQ:
@@ -2023,10 +2041,46 @@ func (c *ctxt0) oprrr(a obj.As) uint32 {
 
 func (c *ctxt0) oprr(a obj.As) uint32 {
 	switch a {
-	case ACLO:
-		return 0x4 << 10
-	case ACLZ:
-		return 0x5 << 10
+	case ACLOW:
+		return 0x4 << 10 // clo.w
+	case ACLZW:
+		return 0x5 << 10 // clz.w
+	case ACTOW:
+		return 0x6 << 10 // cto.w
+	case ACTZW:
+		return 0x7 << 10 // ctz.w
+	case ACLOV:
+		return 0x8 << 10 // clo.d
+	case ACLZV:
+		return 0x9 << 10 // clz.d
+	case ACTOV:
+		return 0xa << 10 // cto.d
+	case ACTZV:
+		return 0xb << 10 // ctz.d
+	case AREVB2H:
+		return 0xc << 10 // revb.2h
+	case AREVB4H:
+		return 0xd << 10 // revb.4h
+	case AREVB2W:
+		return 0xe << 10 // revb.2w
+	case AREVBV:
+		return 0xf << 10 // revb.d
+	case AREVH2W:
+		return 0x10 << 10 // revh.2w
+	case AREVHV:
+		return 0x11 << 10 // revh.d
+	case ABITREV4B:
+		return 0x12 << 10 // bitrev.4b
+	case ABITREV8B:
+		return 0x13 << 10 // bitrev.8b
+	case ABITREVW:
+		return 0x14 << 10 // bitrev.w
+	case ABITREVV:
+		return 0x15 << 10 // bitrev.d
+	case AEXTWH:
+		return 0x16 << 10 // ext.w.h
+	case AEXTWB:
+		return 0x17 << 10 // ext.w.h
 	case ACPUCFG:
 		return 0x1b << 10
 	case ARDTIMELW: