fix(asm): make arm64 and loong64 relocations match the toolchain

asm/loong64_assemble.go

@@ -13,14 +13,17 @@ import (
 
 // assembleLOONG64 assembles a LoongArch (loong64) TEXT function body into
 // machine code.  Every instruction is 4 bytes; the MOV pseudo-instruction and
-// the immediate-arithmetic forms expand to 2–5 instructions when the
+// the immediate-arithmetic forms expand to 2-5 instructions when the
 // immediate does not fit, so the layout is computed in two passes (sizes,
 // then encoding with resolved branch targets).
 //
 // The emitted bytes match the Go toolchain's loong64 assembler, which is the
 // ground-truth oracle: prologue/epilogue, FP/SP frame mapping, branch
 // encodings and the MOV immediate expansions all follow cmd/internal/obj/
-// loong64's asmout cases.
+// loong64's asmout cases.  One deliberate difference: the stack-growth guard
+// (the morestack check in the prologue and the call back into the runtime in
+// the epilogue) is not emitted, so the bytes match only for NOSPLIT functions
+// or zero-frame leaves, where the toolchain emits no guard either.
 func assembleLOONG64(t *ast.Text) ([]byte, map[string]int, []Reloc, []LineEntry, []SpadjStep, error) {
 	fi := loong64ComputeFrame(t)
 	prologue := loong64Prologue(fi)
@@ -69,7 +72,12 @@ func assembleLOONG64(t *ast.Text) ([]byte, map[string]int, []Reloc, []LineEntry,
 			return nil, nil, nil, nil, nil, fmt.Errorf("%s: %w", in.Mnemonic.Text, err)
 		}
 		for j := preCount; j < len(relocs); j++ {
-			relocs[j].Off += pc - len(prologue)
+			// Make the relocation offsets function-relative: each instruction
+			// records its reloc offset relative to its own start, and pc is
+			// that instruction's offset from the function start (prologue
+			// included).  After shifts by the same amount.
+			relocs[j].Off += pc
+			relocs[j].After += pc
 		}
 		preCount = len(relocs)
 		lines = append(lines, LineEntry{Offset: pc, Line: in.Pos().Line})
@@ -417,7 +425,7 @@ func encodeLOONG64Instr(instr *ast.Instr, pc int, offsets map[string]int, fi loo
 
 	case l64Firrr:
 		// ALSL: INSTR $sa, rj, rk, rd (the toolchain's optab places rj in
-		// the second register position); the source amount is 1–4, encoded
+		// the second register position); the source amount is 1-4, encoded
 		// as sa-1.
 		if len(ops) != 4 {
 			return nil, fmt.Errorf("%s expects 4 operands, got %d", mnem, len(ops))
@@ -578,8 +586,8 @@ func encodeLOONG64Branch16(mnem string, op uint32, ops []*ast.Operand, pc int, o
 
 // encodeLOONG64Branch21 encodes a single-register branch: BLTZ/BGEZ and
 // BFPT/BFPF use the 21-bit offset form (register in the rj field), while
-// BGTZ/BLEZ — which the toolchain encodes with the register in the rd field
-// and a 16-bit offset — are handled separately.
+// BGTZ/BLEZ, which the toolchain encodes with the register in the rd field
+// and a 16-bit offset, are handled separately.
 func encodeLOONG64Branch21(mnem string, op uint32, ops []*ast.Operand, pc int, offsets map[string]int, resolve func(string) string) ([]byte, error) {
 	if len(ops) != 2 {
 		return nil, fmt.Errorf("%s expects 2 operands, got %d", mnem, len(ops))
@@ -692,7 +700,7 @@ func encodeLOONG64ImmArith(mnem string, de l64DualEnc, ops []*ast.Operand) ([]by
 }
 
 // isLoong64ShiftD reports whether a shift-immediate opcode constant is one of
-// the 6-bit (.d) variants — the toolchain distinguishes them by the bit
+// the 6-bit (.d) variants, the toolchain distinguishes them by the bit
 // position of the opcode field (bits [25:16]).
 func isLoong64ShiftD(op uint32) bool {
 	return op&0x03ff0000 != 0 && op>>25 == 0
@@ -740,7 +748,7 @@ func l64MemOperands(ops []*ast.Operand, fi loong64FrameInfo) (rd, rj int, off in
 
 // ---- the MOV pseudo-instruction ----
 
-// encodeLOONG64Mov encodes the MOV family — the load/store/immediate
+// encodeLOONG64Mov encodes the MOV family, the load/store/immediate
 // workhorse of Go's loong64 assembly.  MOV is an alias of MOVV (the width
 // mnemonics MOVB/MOVH/MOVW/MOVV/MOVBU/MOVHU/MOVWU/MOVF/MOVD select the
 // access width).  The forms, mirroring the toolchain: