cmd/internal/obj/loong64: add ADDU16I.D instruction support

Go asm syntax:
	ADDV16		$(1<<16), R4, R5

Equivalent platform assembler syntax:
	addu16i.d	r5, r4, $1

Change-Id: Ica4a4e779d0a107cda3eade86027abd6458779a4
Reviewed-on: https://go-review.googlesource.com/c/go/+/699056
Reviewed-by: abner chenc <[email protected]>
Reviewed-by: Michael Pratt <[email protected]>
Auto-Submit: Michael Pratt <[email protected]>
LUCI-TryBot-Result: Go LUCI <[email protected]>
Reviewed-by: Cherry Mui <[email protected]>

Author: sophie-zhao

src/cmd/asm/internal/asm/testdata/loong64enc1.s

@@ -282,6 +282,13 @@ lable2:
 	MOVVP	4(R5), R4		// a4040026
 	MOVVP	(R5), R4		// a4000026
 
+	// ADDU16I.D instruction
+	ADDV16	$(-32768<<16), R4, R5	// ADDV16	$-2147483648, R4, R5	// 85000012
+	ADDV16	$(0<<16), R4, R5	// ADDV16	$0, R4, R5		// 85000010
+	ADDV16	$(8<<16), R4, R5	// ADDV16	$524288, R4, R5		// 85200010
+	ADDV16	$(32767<<16), R4, R5	// ADDV16	$2147418112, R4, R5	// 85fcff11
+	ADDV16	$(16<<16), R4		// ADDV16	$1048576, R4		// 84400010
+
 	// Loong64 atomic memory access instructions
 	AMSWAPB		R14, (R13), R12 // ac395c38
 	AMSWAPH		R14, (R13), R12 // acb95c38

src/cmd/asm/internal/asm/testdata/loong64error.s

@@ -5,3 +5,5 @@
 TEXT errors(SB),$0
 	VSHUF4IV	$16, V1, V2	// ERROR "operand out of range 0 to 15"
 	XVSHUF4IV	$16, X1, X2	// ERROR "operand out of range 0 to 15"
+	ADDV16		$1, R4, R5	// ERROR "the constant must be a multiple of 65536."
+	ADDV16		$65535, R4, R5	// ERROR "the constant must be a multiple of 65536."

src/cmd/internal/obj/loong64/a.out.go

@@ -565,6 +565,9 @@ const (
 	AMOVVF
 	AMOVVD
 
+	// 2.2.1.2
+	AADDV16
+
 	// 2.2.1.3
 	AALSLW
 	AALSLWU

src/cmd/internal/obj/loong64/anames.go

@@ -267,6 +267,9 @@ var optab = []Optab{
 	{AADDV, C_U12CON, C_REG, C_NONE, C_REG, C_NONE, 10, 8, 0, 0},
 	{AADDV, C_U12CON, C_NONE, C_NONE, C_REG, C_NONE, 10, 8, 0, 0},
 
+	{AADDV16, C_32CON, C_REG, C_NONE, C_REG, C_NONE, 4, 4, 0, 0},
+	{AADDV16, C_32CON, C_NONE, C_NONE, C_REG, C_NONE, 4, 4, 0, 0},
+
 	{AAND, C_UU12CON, C_REG, C_NONE, C_REG, C_NONE, 4, 4, 0, 0},
 	{AAND, C_UU12CON, C_NONE, C_NONE, C_REG, C_NONE, 4, 4, 0, 0},
 	{AAND, C_S12CON, C_REG, C_NONE, C_REG, C_NONE, 10, 8, 0, 0},
@@ -1520,6 +1523,7 @@ func buildop(ctxt *obj.Link) {
 			APRELD,
 			APRELDX,
 			AFSEL,
+			AADDV16,
 			obj.ANOP,
 			obj.ATEXT,
 			obj.AFUNCDATA,
@@ -2087,7 +2091,14 @@ func (c *ctxt0) asmout(p *obj.Prog, o *Optab, out []uint32) {
 		if r == 0 {
 			r = int(p.To.Reg)
 		}
-		o1 = OP_12IRR(c.opirr(p.As), uint32(v), uint32(r), uint32(p.To.Reg))
+		if p.As == AADDV16 {
+			if v&65535 != 0 {
+				c.ctxt.Diag("%v: the constant must be a multiple of 65536.\n", p)
+			}
+			o1 = OP_16IRR(c.opirr(p.As), uint32(v>>16), uint32(r), uint32(p.To.Reg))
+		} else {
+			o1 = OP_12IRR(c.opirr(p.As), uint32(v), uint32(r), uint32(p.To.Reg))
+		}
 
 	case 5: // syscall
 		v := c.regoff(&p.From)
@@ -4033,6 +4044,8 @@ func (c *ctxt0) opirr(a obj.As) uint32 {
 		return 0x00b << 22
 	case AADDVU:
 		return 0x00b << 22
+	case AADDV16:
+		return 0x4 << 26
 
 	case AJMP:
 		return 0x14 << 26

src/cmd/internal/obj/loong64/asm.go

@@ -326,6 +326,18 @@ Note: In the following sections 3.1 to 3.6, "ui4" (4-bit unsigned int immediate)
             Go assembly      |      platform assembly
          MOVWP  8(R4), R5    |      ldptr.w r5, r4, $2
 
+6. Note of special add instrction
+    Mapping between Go and platform assembly:
+              Go assembly        |      platform assembly
+      ADDV16  si16<<16, Rj, Rd   |    addu16i.d  rd, rj, si16
+
+      note: si16 is a 16-bit immediate number, and si16<<16 is the actual operand.
+
+    The addu16i.d instruction logically left-shifts the 16-bit immediate number si16 by 16 bits, then
+    sign-extends it. The resulting data is added to the [63:0] bits of data in the general-purpose register
+    rj, and the sum is written into the general-purpose register rd.
+    The addu16i.d instruction is used in conjunction with the ldptr.w/d and stptr.w/d instructions to
+    accelerate access based on the GOT table in position-independent code.
 */
 
 package loong64