From: Guoqi Chen <chenguoqi@loongson.cn>
Date: Thu, 17 Oct 2024 08:26:52 +0000 (+0800)
Subject: cmd/internal/obj/loong64: add {V,XV}PCNT.{B,H,W,D} instructions support
X-Git-Tag: go1.24rc1~450
X-Git-Url: http://www.git.cypherpunks.su/?a=commitdiff_plain;h=2751443e92d9ac6adee2ce0433253fd7b2c04b0f;p=gostls13.git

cmd/internal/obj/loong64: add {V,XV}PCNT.{B,H,W,D} instructions support

Go asm syntax:
          VPCNT{B,H,W,V}  VJ, VD
         XVPCNT{B,H,W,V}  XJ, XD

Equivalent platform assembler syntax:
          vpcnt.{b,w,h,d}  vd, vj
         xvpcnt.{b,w,h,d}  xd, xj

Change-Id: Icec4446b1925745bc3a0bc3f6397d862953b9098
Reviewed-on: https://go-review.googlesource.com/c/go/+/620736
Reviewed-by: Meidan Li <limeidan@loongson.cn>
Reviewed-by: David Chase <drchase@google.com>
Reviewed-by: Cherry Mui <cherryyz@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: sophie zhao <zhaoxiaolin@loongson.cn>
---

diff --git a/src/cmd/asm/internal/asm/testdata/loong64enc1.s b/src/cmd/asm/internal/asm/testdata/loong64enc1.s
index fbfe90a2b0..459d161748 100644
--- a/src/cmd/asm/internal/asm/testdata/loong64enc1.s
+++ b/src/cmd/asm/internal/asm/testdata/loong64enc1.s
@@ -443,3 +443,13 @@ lable2:
 	XVSEQH		X3, X2, X4      // 448c0074
 	XVSEQW		X3, X2, X4      // 440c0174
 	XVSEQV		X3, X2, X4      // 448c0174
+
+	// VPCNT{B,H,W,V}, XVPCNT{B,H,W,V} instruction
+	VPCNTB		V1, V2          // 22209c72
+	VPCNTH		V1, V2          // 22249c72
+	VPCNTW		V1, V2          // 22289c72
+	VPCNTV		V1, V2          // 222c9c72
+	XVPCNTB		X3, X2          // 62209c76
+	XVPCNTH		X3, X2          // 62249c76
+	XVPCNTW		X3, X2          // 62289c76
+	XVPCNTV		X3, X2          // 622c9c76
diff --git a/src/cmd/internal/obj/loong64/a.out.go b/src/cmd/internal/obj/loong64/a.out.go
index 4e4135b017..80fba3c7b7 100644
--- a/src/cmd/internal/obj/loong64/a.out.go
+++ b/src/cmd/internal/obj/loong64/a.out.go
@@ -628,6 +628,16 @@ const (
 	AVMOVQ
 	AXVMOVQ
 
+	// LSX and LASX Bit-manipulation Instructions
+	AVPCNTB
+	AVPCNTH
+	AVPCNTW
+	AVPCNTV
+	AXVPCNTB
+	AXVPCNTH
+	AXVPCNTW
+	AXVPCNTV
+
 	// LSX and LASX integer comparison instruction
 	AVSEQB
 	AXVSEQB
diff --git a/src/cmd/internal/obj/loong64/anames.go b/src/cmd/internal/obj/loong64/anames.go
index a316c75522..3d2f329917 100644
--- a/src/cmd/internal/obj/loong64/anames.go
+++ b/src/cmd/internal/obj/loong64/anames.go
@@ -257,6 +257,14 @@ var Anames = []string{
 	"FTINTRNEVD",
 	"VMOVQ",
 	"XVMOVQ",
+	"VPCNTB",
+	"VPCNTH",
+	"VPCNTW",
+	"VPCNTV",
+	"XVPCNTB",
+	"XVPCNTH",
+	"XVPCNTW",
+	"XVPCNTV",
 	"VSEQB",
 	"XVSEQB",
 	"VSEQH",
diff --git a/src/cmd/internal/obj/loong64/asm.go b/src/cmd/internal/obj/loong64/asm.go
index 48a9a5389f..d1200fa598 100644
--- a/src/cmd/internal/obj/loong64/asm.go
+++ b/src/cmd/internal/obj/loong64/asm.go
@@ -93,6 +93,8 @@ var optab = []Optab{
 	{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},
 	{AMOVD, C_FREG, C_NONE, C_NONE, C_FREG, C_NONE, 9, 4, 0, 0},
+	{AVPCNTB, C_VREG, C_NONE, C_NONE, C_VREG, C_NONE, 9, 4, 0, 0},
+	{AXVPCNTB, C_XREG, C_NONE, C_NONE, C_XREG, C_NONE, 9, 4, 0, 0},
 
 	{AFMADDF, C_FREG, C_FREG, C_NONE, C_FREG, C_NONE, 39, 4, 0, 0},
 	{AFMADDF, C_FREG, C_FREG, C_FREG, C_FREG, C_NONE, 39, 4, 0, 0},
@@ -1294,6 +1296,16 @@ func buildop(ctxt *obj.Link) {
 			opset(AXVSEQH, r0)
 			opset(AXVSEQW, r0)
 			opset(AXVSEQV, r0)
+
+		case AVPCNTB:
+			opset(AVPCNTH, r0)
+			opset(AVPCNTW, r0)
+			opset(AVPCNTV, r0)
+
+		case AXVPCNTB:
+			opset(AXVPCNTH, r0)
+			opset(AXVPCNTW, r0)
+			opset(AXVPCNTV, r0)
 		}
 	}
 }
@@ -2322,6 +2334,22 @@ func (c *ctxt0) oprr(a obj.As) uint32 {
 		return 0x46b9 << 10 // ftintrne.l.s
 	case AFTINTRNEVD:
 		return 0x46ba << 10 // ftintrne.l.d
+	case AVPCNTB:
+		return 0x1ca708 << 10 // vpcnt.b
+	case AVPCNTH:
+		return 0x1ca709 << 10 // vpcnt.h
+	case AVPCNTW:
+		return 0x1ca70a << 10 // vpcnt.w
+	case AVPCNTV:
+		return 0x1ca70b << 10 // vpcnt.v
+	case AXVPCNTB:
+		return 0x1da708 << 10 // xvpcnt.b
+	case AXVPCNTH:
+		return 0x1da709 << 10 // xvpcnt.h
+	case AXVPCNTW:
+		return 0x1da70a << 10 // xvpcnt.w
+	case AXVPCNTV:
+		return 0x1da70b << 10 // xvpcnt.v
 	}
 
 	c.ctxt.Diag("bad rr opcode %v", a)