fix(asm): validate the loong64 ll/sc offset span like the toolchain

Assisted-by: GLM 5.3 Flash

asm/loong64_more_test.go

@@ -131,6 +131,98 @@ TEXT ·ptr(SB), NOSPLIT, $0
 	)
 }
 
+// TestLOONG64_firr14Spans pins the 2RI14 offset spans of the LL/SC/MOVWP
+// families against the toolchain's operand classes: an unaligned offset is
+// rejected ("offset must be a multiple of 4"), a signed 16-bit offset rides
+// in si14 alone, and anything wider materialises its high half in R30 with
+// addu16i.d before the base folds in, with si14 keeping bits 15:2.  The
+// pinned words are GOARCH=loong64 go tool asm's own bytes for the same
+// sources, boundaries included (-32766 and 32766 are the class edges).
+func TestLOONG64_firr14Spans(t *testing.T) {
+	fn := firstTextLOONG64(t, `#include "textflag.h"
+TEXT ·spans(SB), NOSPLIT, $0
+	SC R4, 32764(R5)
+	SC R4, -32764(R5)
+	SC R4, 32768(R5)
+	SC R4, -32768(R5)
+	SC R4, -32772(R5)
+	LL 65540(R5), R4
+	LLV 65540(R5), R4
+	SCV R4, 65540(R5)
+	MOVWP R4, 65540(R5)
+	RET
+`)
+	code := assembleLOONG64Helper(t, fn)
+	wantWords(t, code,
+		0x217FFCA4, // sc.w r4, 8191(r5) — inside si14
+		0x218004A4, // sc.w r4, -8191(r5)
+		0x1000001E, // addu16i.d r30, r0, 0
+		0x001097DE, // add.d r30, r30, r5
+		0x218003C4, // sc.w r4, 8192(r30)
+		0x13FFFC1E, // addu16i.d r30, r0, -1
+		0x001097DE, // add.d r30, r30, r5
+		0x218003C4, // sc.w r4, 8192(r30)
+		0x13FFFC1E, // addu16i.d r30, r0, -1
+		0x001097DE, // add.d r30, r30, r5
+		0x217FFFC4, // sc.w r4, 8191(r30)
+		0x1000041E, // addu16i.d r30, r0, 1
+		0x001097DE, // add.d r30, r30, r5
+		0x200007C4, // ll.w r4, 1(r30)
+		0x1000041E, // addu16i.d r30, r0, 1
+		0x001097DE, // add.d r30, r30, r5
+		0x220007C4, // ldptr.d r4, 1(r30)
+		0x1000041E, // addu16i.d r30, r0, 1
+		0x001097DE, // add.d r30, r30, r5
+		0x230007C4, // sc.d r4, 1(r30)
+		0x1000041E, // addu16i.d r30, r0, 1
+		0x001097DE, // add.d r30, r30, r5
+		0x250007C4, // stptr.w r4, 1(r30)
+		0x4C000020,
+	)
+
+	// The expansion participates in layout: the function size carries the
+	// six three-word forms plus the single-word pair and the RET.
+	f, errs := parser.Parse("spans_loong64.s", `#include "textflag.h"
+TEXT ·spans(SB), NOSPLIT, $0
+	SC R4, 32768(R5)
+	LL 65540(R5), R4
+	RET
+`)
+	if len(errs) > 0 {
+		t.Fatalf("parse: %v", errs)
+	}
+	img, err := AssembleFileLOONG64(f)
+	if err != nil {
+		t.Fatalf("AssembleFileLOONG64: %v", err)
+	}
+	if img.Funcs[0].Size != 8*3+4 {
+		t.Errorf("function size = %d, want %d", img.Funcs[0].Size, 8*3+4)
+	}
+}
+
+// TestLOONG64_firr14Alignment checks the LL/SC/MOVWP alignment rule the
+// toolchain enforces as "offset must be a multiple of 4": the LoongArch
+// ll/sc family requires a naturally scaled offset, and si14 cannot express
+// the remainder.  GOARCH=loong64 go tool asm rejects every case here.
+func TestLOONG64_firr14Alignment(t *testing.T) {
+	cases := []string{
+		"SC R4, 1(R5)",
+		"SCW R4, -1(R5)",
+		"SCV R4, 2(R5)",
+		"LL 1(R5), R4",
+		"LLW 3(R5), R4",
+		"LLV 6(R5), R4",
+		"MOVWP R4, 1(R5)",
+		"MOVVP R4, 2(R5)",
+	}
+	for _, src := range cases {
+		fn := firstTextLOONG64(t, "#include \"textflag.h\"\nTEXT ·e(SB), NOSPLIT, $0\n\t"+src+"\n\tRET\n")
+		if _, _, _, _, _, err := assembleLOONG64(fn); err == nil {
+			t.Errorf("%s: expected an error, got none", src)
+		}
+	}
+}
+
 // TestLOONG64_atomics exercises the AM* read-modify-write forms and
 // RDTIME, plus the MOVV FP→GP move.
 func TestLOONG64_atomics(t *testing.T) {