cmd/internal/obj/riscv: add support for vector mask instructions

Add support for vector mask instructions to the RISC-V assembler.
These allow manipulation of vector masks and include mask register
logical instructions, population count and find-first bit set
instructions.

Change-Id: I3ab3aa0f918338aee9b37ac5a2b2fdc407875072
Reviewed-on: https://go-review.googlesource.com/c/go/+/646779
Reviewed-by: Carlos Amedee <carlos@golang.org>
Reviewed-by: Meng Zhuo <mengzhuo1203@gmail.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Mark Ryan <markdryan@rivosinc.com>
Reviewed-by: Junyang Shao <shaojunyang@google.com>

src/cmd/asm/internal/asm/testdata/riscv64.s

@@ -1201,6 +1201,34 @@ start:
 	VFWREDUSUMVS	V1, V2, V3			// d79120c6
 	VFWREDUSUMVS	V1, V2, V0, V3			// d79120c4
 
+	// 31.15: Vector Mask Instructions
+	VMANDMM		V1, V2, V3			// d7a12066
+	VMNANDMM	V1, V2, V3			// d7a12076
+	VMANDNMM	V1, V2, V3			// d7a12062
+	VMXORMM		V1, V2, V3			// d7a1206e
+	VMORMM		V1, V2, V3			// d7a1206a
+	VMNORMM		V1, V2, V3			// d7a1207a
+	VMORNMM		V1, V2, V3			// d7a12072
+	VMXNORMM	V1, V2, V3			// d7a1207e
+	VMMVM		V2, V3				// d7212166
+	VMCLRM		V3				// d7a1316e
+	VMSETM		V3				// d7a1317e
+	VMNOTM		V2, V3				// d7212176
+	VCPOPM		V2, X10				// 57252842
+	VCPOPM		V2, V0, X10			// 57252840
+	VFIRSTM		V2, X10				// 57a52842
+	VFIRSTM		V2, V0, X10			// 57a52840
+	VMSBFM		V2, V3				// d7a12052
+	VMSBFM		V2, V0, V3			// d7a12050
+	VMSIFM		V2, V3				// d7a12152
+	VMSIFM		V2, V0, V3			// d7a12150
+	VMSOFM		V2, V3				// d7212152
+	VMSOFM		V2, V0, V3			// d7212150
+	VIOTAM		V2, V3				// d7212852
+	VIOTAM		V2, V0, V3			// d7212850
+	VIDV		V3				// d7a10852
+	VIDV		V0, V3				// d7a10850
+
 	//
 	// Privileged ISA
 	//

src/cmd/asm/internal/asm/testdata/riscv64error.s

@@ -362,5 +362,11 @@ TEXT errors(SB),$0
 	VFREDMAXVS	V1, V2, V4, V3			// ERROR "invalid vector mask register"
 	VFREDMINVS	V1, V2, V4, V3			// ERROR "invalid vector mask register"
 	VFWREDOSUMVS	V1, V2, V4, V3			// ERROR "invalid vector mask register"
+	VCPOPM		V2, V4, X10			// ERROR "invalid vector mask register"
+	VFIRSTM		V2, V4, X10			// ERROR "invalid vector mask register"
+	VMSBFM		V2, V4, V3			// ERROR "invalid vector mask register"
+	VMSIFM		V2, V4, V3			// ERROR "invalid vector mask register"
+	VMSOFM		V2, V4, V3			// ERROR "invalid vector mask register"
+	VIOTAM		V2, V4, V3			// ERROR "invalid vector mask register"
 
 	RET

src/cmd/asm/internal/asm/testdata/riscv64validation.s

@@ -380,5 +380,24 @@ TEXT validation(SB),$0
 	VFREDMINVS	X10, V2, V3			// ERROR "expected vector register in vs1 position"
 	VFWREDOSUMVS	X10, V2, V3			// ERROR "expected vector register in vs1 position"
 	VFWREDUSUMVS	X10, V2, V3			// ERROR "expected vector register in vs1 position"
+	VMANDMM		X10, V2, V3			// ERROR "expected vector register in vs1 position"
+	VMNANDMM	X10, V2, V3			// ERROR "expected vector register in vs1 position"
+	VMANDNMM	X10, V2, V3			// ERROR "expected vector register in vs1 position"
+	VMXORMM		X10, V2, V3			// ERROR "expected vector register in vs1 position"
+	VMORMM		X10, V2, V3			// ERROR "expected vector register in vs1 position"
+	VMNORMM		X10, V2, V3			// ERROR "expected vector register in vs1 position"
+	VMORNMM		X10, V2, V3			// ERROR "expected vector register in vs1 position"
+	VMXNORMM	X10, V2, V3			// ERROR "expected vector register in vs1 position"
+	VMMVM		V3, X10				// ERROR "expected vector register in vd position"
+	VMNOTM		V3, X10				// ERROR "expected vector register in vd position"
+	VCPOPM		V2, V1				// ERROR "expected integer register in rd position"
+	VCPOPM		X11, X10			// ERROR "expected vector register in vs2 position"
+	VFIRSTM		V2, V1				// ERROR "expected integer register in rd position"
+	VFIRSTM		X11, X10			// ERROR "expected vector register in vs2 position"
+	VMSBFM		X10, V3				// ERROR "expected vector register in vs2 position"
+	VMSIFM		X10, V3				// ERROR "expected vector register in vs2 position"
+	VMSOFM		X10, V3				// ERROR "expected vector register in vs2 position"
+	VIOTAM		X10, V3				// ERROR "expected vector register in vs2 position"
+	VIDV		X10				// ERROR "expected vector register in vd position"
 
 	RET

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

@@ -652,12 +652,16 @@ var Anames = []string{
 	"SNEZ",
 	"VFABSV",
 	"VFNEGV",
-	"VMFGEVV",
-	"VMFGTVV",
 	"VL1RV",
 	"VL2RV",
 	"VL4RV",
 	"VL8RV",
+	"VMCLRM",
+	"VMFGEVV",
+	"VMFGTVV",
+	"VMMVM",
+	"VMNOTM",
+	"VMSETM",
 	"VMSGEUVI",
 	"VMSGEUVV",
 	"VMSGEVI",

src/cmd/internal/obj/riscv/cpu.go

@@ -1180,12 +1180,16 @@ const (
 	ASNEZ
 	AVFABSV
 	AVFNEGV
-	AVMFGEVV
-	AVMFGTVV
 	AVL1RV
 	AVL2RV
 	AVL4RV
 	AVL8RV
+	AVMCLRM
+	AVMFGEVV
+	AVMFGTVV
+	AVMMVM
+	AVMNOTM
+	AVMSETM
 	AVMSGEUVI
 	AVMSGEUVV
 	AVMSGEVI

src/cmd/internal/obj/riscv/obj.go

@@ -1328,6 +1328,13 @@ func validateRVFV(ctxt *obj.Link, ins *instruction) {
 	wantNoneReg(ctxt, ins, "rs3", ins.rs3)
 }
 
+func validateRVI(ctxt *obj.Link, ins *instruction) {
+	wantIntReg(ctxt, ins, "rd", ins.rd)
+	wantNoneReg(ctxt, ins, "rs1", ins.rs1)
+	wantVectorReg(ctxt, ins, "vs2", ins.rs2)
+	wantNoneReg(ctxt, ins, "rs3", ins.rs3)
+}
+
 func validateRVIV(ctxt *obj.Link, ins *instruction) {
 	wantVectorReg(ctxt, ins, "vd", ins.rd)
 	wantIntReg(ctxt, ins, "rs1", ins.rs1)
@@ -1577,6 +1584,10 @@ func encodeRVFV(ins *instruction) uint32 {
 	return encodeR(ins.as, regF(ins.rs1), regV(ins.rs2), regV(ins.rd), ins.funct3, ins.funct7)
 }
 
+func encodeRVI(ins *instruction) uint32 {
+	return encodeR(ins.as, 0, regV(ins.rs2), regI(ins.rd), ins.funct3, ins.funct7)
+}
+
 func encodeRVIV(ins *instruction) uint32 {
 	return encodeR(ins.as, regI(ins.rs1), regV(ins.rs2), regV(ins.rd), ins.funct3, ins.funct7)
 }
@@ -1881,6 +1892,7 @@ var (
 	rIFEncoding   = encoding{encode: encodeRIF, validate: validateRIF, length: 4}
 	rFFEncoding   = encoding{encode: encodeRFF, validate: validateRFF, length: 4}
 	rVFVEncoding  = encoding{encode: encodeRVFV, validate: validateRVFV, length: 4}
+	rVIEncoding   = encoding{encode: encodeRVI, validate: validateRVI, length: 4}
 	rVIVEncoding  = encoding{encode: encodeRVIV, validate: validateRVIV, length: 4}
 	rVVEncoding   = encoding{encode: encodeRVV, validate: validateRVV, length: 4}
 	rVViEncoding  = encoding{encode: encodeRVVi, validate: validateRVVi, length: 4}
@@ -2609,6 +2621,23 @@ var instructions = [ALAST & obj.AMask]instructionData{
 	AVFWREDOSUMVS & obj.AMask: {enc: rVVVEncoding},
 	AVFWREDUSUMVS & obj.AMask: {enc: rVVVEncoding},
 
+	// 31.15: Vector Mask Instructions
+	AVMANDMM & obj.AMask:  {enc: rVVVEncoding},
+	AVMNANDMM & obj.AMask: {enc: rVVVEncoding},
+	AVMANDNMM & obj.AMask: {enc: rVVVEncoding},
+	AVMXORMM & obj.AMask:  {enc: rVVVEncoding},
+	AVMORMM & obj.AMask:   {enc: rVVVEncoding},
+	AVMNORMM & obj.AMask:  {enc: rVVVEncoding},
+	AVMORNMM & obj.AMask:  {enc: rVVVEncoding},
+	AVMXNORMM & obj.AMask: {enc: rVVVEncoding},
+	AVCPOPM & obj.AMask:   {enc: rVIEncoding},
+	AVFIRSTM & obj.AMask:  {enc: rVIEncoding},
+	AVMSBFM & obj.AMask:   {enc: rVVEncoding},
+	AVMSIFM & obj.AMask:   {enc: rVVEncoding},
+	AVMSOFM & obj.AMask:   {enc: rVVEncoding},
+	AVIOTAM & obj.AMask:   {enc: rVVEncoding},
+	AVIDV & obj.AMask:     {enc: rVVEncoding},
+
 	//
 	// Privileged ISA
 	//
@@ -3765,6 +3794,47 @@ func instructionsForProg(p *obj.Prog) []*instruction {
 			ins.as = AVFSGNJNVV
 		}
 		ins.rd, ins.rs1, ins.rs2 = uint32(p.To.Reg), uint32(p.From.Reg), uint32(p.From.Reg)
+
+	case AVMANDMM, AVMNANDMM, AVMANDNMM, AVMXORMM, AVMORMM, AVMNORMM, AVMORNMM, AVMXNORMM, AVMMVM, AVMNOTM:
+		ins.rd, ins.rs1, ins.rs2 = uint32(p.To.Reg), uint32(p.From.Reg), uint32(p.Reg)
+		switch ins.as {
+		case AVMMVM:
+			ins.as, ins.rs2 = AVMANDMM, ins.rs1
+		case AVMNOTM:
+			ins.as, ins.rs2 = AVMNANDMM, ins.rs1
+		}
+
+	case AVMCLRM, AVMSETM:
+		ins.rd, ins.rs1, ins.rs2 = uint32(p.From.Reg), uint32(p.From.Reg), uint32(p.From.Reg)
+		switch ins.as {
+		case AVMCLRM:
+			ins.as = AVMXORMM
+		case AVMSETM:
+			ins.as = AVMXNORMM
+		}
+
+	case AVCPOPM, AVFIRSTM, AVMSBFM, AVMSIFM, AVMSOFM, AVIOTAM:
+		// Set mask bit
+		switch {
+		case ins.rs1 == obj.REG_NONE:
+			ins.funct7 |= 1 // unmasked
+		case ins.rs1 != REG_V0:
+			p.Ctxt.Diag("%v: invalid vector mask register", p)
+		}
+		ins.rs1 = obj.REG_NONE
+
+	case AVIDV:
+		// Set mask bit
+		switch {
+		case ins.rd == obj.REG_NONE:
+			ins.funct7 |= 1 // unmasked
+		case ins.rd != obj.REG_NONE && ins.rs2 != REG_V0:
+			p.Ctxt.Diag("%v: invalid vector mask register", p)
+		}
+		if ins.rd == obj.REG_NONE {
+			ins.rd = uint32(p.From.Reg)
+		}
+		ins.rs1, ins.rs2 = obj.REG_NONE, REG_V0
 	}
 
 	for _, ins := range inss {