build: rename the module to sourcedock.dev/petrbalvin/gasm-sdk

Assisted-by: GLM 5.3 Flash

.gitea/workflows/release.yml

@@ -342,7 +342,10 @@ jobs:
               my @cmd = (q{curl}, q{-sS}, q{-o}, q{/dev/null}, q{-w}, q{%{http_code}},
                 q{-H}, qq{Authorization: token $ENV{GITEA_TOKEN}},
                 q{-H}, q{Content-Type: application/octet-stream},
-                q{-X}, q{POST}, q{--data-binary}, qq{@$path},
+                # The @ must not sit inside a qq{} string: there it starts an
+                # array interpolation and the upload body collapses to empty,
+                # which Gitea stores as a 201-created zero-byte attachment.
+                q{-X}, q{POST}, q{--data-binary}, q{@} . $path,
                 qq{$ENV{GITEA_SERVER_URL}/api/v1/repos/$ENV{GITEA_REPOSITORY}/releases/$id/assets?name=$name});
               open(my $curl, q{-|}, @cmd) or die qq{curl: $!};
               my $code = <$curl>;

.gitea/workflows/test.yml

@@ -56,6 +56,28 @@ jobs:
       - name: Build
         run: go build ./...
 
+      - name: FreeBSD build (amd64)
+        # The debugger's ptrace surface and the JIT substrate are the two
+        # FreeBSD-portable layers the tree carries; the forge has no FreeBSD
+        # runner, so a push can only compile-gate them. Running the ptrace
+        # suite needs real FreeBSD hardware.
+        env:
+          GOOS: freebsd
+          GOARCH: amd64
+        run: go build ./...
+
+      - name: FreeBSD build (arm64)
+        env:
+          GOOS: freebsd
+          GOARCH: arm64
+        run: go build ./...
+
+      - name: FreeBSD build (riscv64)
+        env:
+          GOOS: freebsd
+          GOARCH: riscv64
+        run: go build ./...
+
       - name: Format
         run: |
           perl -e '

CHANGELOG.md

@@ -1,6 +1,6 @@
 # Changelog
 
-All notable changes to gasm-devkit are documented here.
+All notable changes to gasm-sdk are documented here.
 
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
@@ -9,14 +9,87 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ### Added
 
-- **GOOBJ format specification.**  [docs/GOOBJ.md](docs/GOOBJ.md)
-  documents the Go object file format in full: both containers, the 96
-  byte header and all 19 blocks, every structure with its byte
-  offsets, symbol kinds and flag bits, all 106 relocation types with
-  the weak variants, aux symbols, the FuncInfo payload, the pc-value
-  table encoding, the content hashes and the builtin table, all
-  verified byte for byte against objects produced by Go 1.27.1's own
-  tools.
+- **The FreeBSD port of the debugger.**  `gasm debug` runs on FreeBSD on
+  amd64, arm64 and riscv64: the same interactive surface as on Linux —
+  breakpoints, hardware watchpoints (x86 debug registers, the arm64 debug
+  register file), single-stepping, register and memory access — behind the
+  kernel's own ptrace requests, with tracee memory through `PT_IO` and
+  stop reports through `PT_LWPINFO`.  The JIT substrate maps executable
+  memory through `golang.org/x/sys/unix`, so `verify` builds on FreeBSD
+  too.  The pipeline compile-gates all three architectures; live
+  validation awaits a FreeBSD machine.
+- **Workspace-wide navigation in the language server.**  `gasm lsp` indexes
+  the `.s` files under the workspace root beyond the documents the editor
+  has open, so go-to-definition, find references and workspace symbol search
+  reach files that were never opened.  An open buffer always shadows its
+  disk copy, and watched-file events together with a per-query freshness
+  check keep the index current.
+- **Quick fixes for the textflag include and the argument area.**  The
+  `missing-textflag-include` warning offers to add the include after the
+  last one in the file, and the `abi-argsize` warning offers to set the
+  TEXT argument area to the size the `// func` signature implies, computed
+  by the new `lint.ExpectedArgSize`.
+
+### Changed
+
+- **The corpus audit assembles like the build.**  A file's `//go:build`
+  constraint decides which target architectures attempt it: cpu_x86.s is
+  an x86 build alone, and the msan and goexperiment.runtimesecret trees
+  are compiled by no supported build, so they leave the measured set
+  instead of failing it.  The headline now reads "assemble for every
+  applicable target": every real-code GOROOT assembly file, the tree
+  without testdata, assembles for all four architectures (250 of 250,
+  100 %); over the whole tree including testdata the measure is 271 of
+  322 (84.2 %).
+- **The module moves to `sourcedock.dev/petrbalvin/gasm-sdk`.**  The
+  repository and the module rename together with the product, now the
+  GAsm Software Development Kit.  Fresh installs become
+  `go install sourcedock.dev/petrbalvin/gasm-sdk/cmd/gasm@latest`, and
+  installs pinned to the old `gasm-sdk` path stop resolving once the
+  repository takes the new name: reinstall from the new path.  The
+  binary stays `gasm`.
+
+### Fixed
+
+- **Rename edits land in their own documents.**  A rename collected the
+  ranges of every reference across the open documents but applied them all
+  to the document that started it, so renaming a symbol used in a second
+  file moved that file's text into the first.  Each edit now applies to the
+  document it was collected in.
+- **Negative numeric PC-relative jumps.**  `JMP -3(PC)`, the shape the
+  runtime's exit loops write (sys_linux_amd64.s, sys_netbsd_amd64.s),
+  resolved to nothing: only the forward forms counted.  A negative count
+  now walks the same instruction statements backwards, labels excluded,
+  byte-identical with the toolchain.
+- **The arm64 move-wide family reads its immediate as an unsigned
+  pattern.**  `MOVK $(40000<<48)` folds to a negative int64 and was
+  rejected; the toolchain picks the 16-bit lane from the 64-bit bit
+  pattern, so the encoder now does the same, and a zero immediate is
+  rejected where the toolchain rejects it.
+
+## [0.35.0] - 2026-09-22
+
+### Added
+
+- **The go_asm.h generator.**  `gasm asm` generates the package's go_asm.h
+  itself when an assembly file includes it: the Go files beside the source
+  are type-checked for the target architecture and the constants and field
+  offsets become assembler defines, so package-context files assemble with
+  no compiler and no `go build` in the loop.  `-GOOS` selects the
+  type-checking GOOS for GOOS-specific files, and the corpus audit derives
+  the GOOS from the file name.
+- **ELF data relocations on arm64, riscv64 and loong64.**  `gasm asm
+  --format elf` emits `.rela.data` for symbol-valued DATA initialisers on
+  every architecture (amd64 carried them already), so standalone ELF
+  objects link on all four targets.
+- **Corpus failure listing.**  `gasm audit-instructions --corpus --list`
+  prints every failing file with its failure reason, per architecture,
+  instead of one representative file per reason.
+- **DATA with symbol values and relaxed symbol spellings.**  DATA
+  initialisers accept `$symbol(SB)` values, laid down as an absolute
+  relocation at the data field (GOOBJ on all four architectures and ELF
+  on all four as of this release), and U+2215 is accepted inside symbol
+  package paths.
 - **Macro expansion and include splicing.**  `gasm asm`, `gasm diff` and
   `gasm audit-instructions` now preprocess assembly the way the
   toolchain does: object and parameterised `#define` macros expand at
@@ -57,21 +130,77 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
   ranges, index-only VSIB memory operands and bare trailing immediates;
   macro substitution reaches parameters used with element suffixes
   (`A.S4`), and `;` separates statements in plain files.
-- **`gasm asm -GOOS`.**  The go_asm.h generator type-checks per target
-  GOOS, so the darwin-only and windows-only runtime files assemble with
-  their own defines; the corpus audit derives the GOOS from the file
-  name.  DATA initialisers accept `$symbol(SB)` values (an absolute
-  relocation at the data field, GOOBJ on all four architectures and ELF
-  on amd64, arm64, riscv64 and loong64), and U+2215 is accepted inside
-  symbol package paths.
-- **The corpus audit measures honestly.**  Files named for Go ports gasm
-  does not target (arm, 386, s390x, ...) are no longer attempted for the
-  four supported architectures (no supported build compiles them), and
-  the headline rate is reported over attemptable files: 136 of 433 on
-  the full corpus (31.4 %), 135 of 383 on real code (35.2 %), from the
-  127 that the previous release measured.  The probe battery that
-  decides encodability gained the operand shapes the new families use.
--
+- **Per-architecture reference pages.**  [docs/asm/](docs/asm/README.md)
+  gains AMD64, ARM64, RISCV64 and LOONG64: the register files and the
+  roles the ABI fixes, addressing, operand order with every special form,
+  constants and materialisation, alignment, fences and the relocations
+  each target emits.  An instruction inventory appendix per architecture
+  is generated from the toolchain's own tables by `just gen`, and the
+  regenerated tables recognise 147 more mnemonics than the previous
+  release carried (arm64 107, riscv64 31, loong64 9).
+- **The Plan 9 assembly language reference.**  [docs/asm/](docs/asm/README.md)
+  opens the complete language reference with its common core: the lexicon,
+  statement structure and constant expressions, the operand grammar with
+  the pseudo-registers and symbol naming, the directives and the function
+  flag vocabulary, preprocessing with `#define` and `#include`, and the
+  Go-embedded layer (ABI0, prototypes, `go_asm.h`, `funcdata.h` and the
+  runtime contract).  Every claim is verified against `go tool asm` of
+  Go 1.27.1 and gasm's differential tests; the per-architecture pages and
+  generated instruction appendices follow.
+- **GOOBJ format specification.**  [docs/GOOBJ.md](docs/GOOBJ.md)
+  documents the Go object file format in full: both containers, the 96
+  byte header and all 19 blocks, every structure with its byte
+  offsets, symbol kinds and flag bits, all 106 relocation types with
+  the weak variants, aux symbols, the FuncInfo payload, the pc-value
+  table encoding, the content hashes and the builtin table, all
+  verified byte for byte against objects produced by Go 1.27.1's own
+  tools.
+
+### Changed
+
+- **The corpus audit measures like a build.**  Files named for a Go port
+  gasm does not target (arm, 386, s390x, ...) are never attempted, because
+  no supported build compiles them; the GOOS comes from the file name; and
+  each target's go_asm.h is generated on the fly.  The headline is reported
+  over attemptable files: 291 of 353 on the full corpus (82.4 %) assemble
+  for every target architecture and 295 of 303 on real code (97.4 %),
+  against 108 of 627 over all files (17.2 %) that the previous release
+  measured.
+
+### Fixed
+
+- **The operand forms GOROOT writes.**  Numeric PC-relative jumps
+  (`JEQ 2(PC)`, the park loop `JMP 0(PC)`) resolve with the toolchain's
+  own instruction counting and fold jump-to-jump chains exactly as its
+  branch optimiser does; symbol immediates (`MOVQ $sym(SB), AX`)
+  assemble to the toolchain's RIP-relative LEA with an R_PCREL
+  relocation; negated constant expressions in operands (`ADJSP
+  $-(REGS - 8)`, the shape the cgo ABI macros write) fold; the immediate
+  multiply (`IMULQ $1000000000, AX`) encodes with the toolchain's
+  0x69/0x6B selection; the TLS access pair assembles as the toolchain's
+  one-instruction form (the bare `MOVQ TLS, r` load nops out and
+  `off(r)(TLS*1)` folds to the segment-prefixed absolute whose disp32
+  carries the R_TLSLE relocation, per-GOOS); arm64 accepts the
+  bare-register indirect branch (`BL R9` beside `BL (R9)`, both BLR) and
+  the zero-immediate store (`MOVD $0, mem` through the zero register,
+  rejecting non-zero immediates as the toolchain does); `PCALIGN` now
+  aligns on amd64, padding with the toolchain's greedy
+  single-instruction NOPs; the segment-absolute forms (`MOVQ 0x30(GS),
+  AX` and the store direction) and the absolute crash-store
+  (`MOVL $0xf1, 0xf1`) encode; and `gasm asm` predefines the
+  `GOARCH_<arch>` and `GOOS_<goos>` macros the go command passes to
+  `go tool asm`, so GOROOT headers' `#ifdef GOARCH_amd64` platform
+  blocks (`go_tls.h`'s `get_tls` and friends) select as intended.  The
+  GOROOT corpus measure moves to 291 of 353 files assembling for every
+  target architecture (82.4 %), 97.4 % of the real-code corpus, from
+  70.8 % and 82.2 %.
+- **Tool corrections across the pipeline.**  The formatter keeps square
+  brackets in SIMD operands, statement separators and canonical macro
+  bodies; the linter drops false positives on shift counts, SETcc
+  spellings and ABIInternal references; the lexer treats a trailing
+  carriage return as a line end so comment text stays idempotent; and
+  arm64 rejects bare BTI with a diagnostic while accepting the full
+  family.
 
 ## [0.34.0] - 2026-09-20
 

CONTRIBUTING.md

@@ -1,6 +1,6 @@
 # Contributing
 
-Contributions to **gasm-devkit** are governed by the Contributor terms
+Contributions to **gasm-sdk** are governed by the Contributor terms
 below; submitting one means you accept them.
 
 ## Contributor terms
@@ -29,8 +29,8 @@ compiler (gcc), because `just gates` includes `just race` and the race
 detector needs cgo.
 
 ```sh
-git clone https://sourcedock.dev/petrbalvin/gasm-devkit.git
-cd gasm-devkit
+git clone https://sourcedock.dev/petrbalvin/gasm-sdk.git
+cd gasm-sdk
 just build
 just gates
 ```
@@ -126,7 +126,7 @@ tag, where it would double the time and the memory a shared runner cannot spare.
 
 ## Reporting bugs
 
-Open an issue at `https://sourcedock.dev/petrbalvin/gasm-devkit/issues` with the
+Open an issue at `https://sourcedock.dev/petrbalvin/gasm-sdk/issues` with the
 version, the operating system and architecture, the exact command, the full output,
 and the expected against the actual behaviour.
 

README.md

@@ -1,6 +1,6 @@
-# Plan 9 assembly tooling, inside and outside Go
+# GAsm: Software Development Kit for Plan 9 Assembly
 
-> **Warning: this is an experiment.** gasm-devkit is under active
+> **Warning: this is an experiment.** gasm-sdk is under active
 > development and is not stable. The version is 0.x.x: commands, flags,
 > output formats and behaviour can change without warning at any time.
 > A 1.0.0 release is light years away. Nothing in this document is a
@@ -13,7 +13,7 @@
 there is no formatter, no linter and no debugger for `.s` files, and no
 assembler that works without a Go installation. Developers write
 assembly blind, validate it by benchmark, and debug it by print
-statement. gasm-devkit is the missing toolkit: a single, self-contained
+statement. gasm-sdk is the missing toolkit: a single, self-contained
 binary, `gasm`, that serves both purposes.
 
 - **Help develop Plan 9 assembly.** Formatting, linting, disassembly,
@@ -58,7 +58,7 @@ Plan 9 (Go):     MOVQ AX, total-16(SP)
 
 The same lines, but only one of them tells you what the number is for.
 The syntax is uppercase, regular and boring, which is the highest
-compliment a language for machine code can earn. gasm-devkit exists
+compliment a language for machine code can earn. gasm-sdk exists
 to give that syntax the tooling it deserves.
 
 ## Features
@@ -81,7 +81,10 @@ to give that syntax the tooling it deserves.
   GOOBJ format, which needs the installed toolchain and which `go build`
   consumes in place of the toolchain's output.  Framed functions get the
   stack-split guard and the morestack block, byte-identical to the
-  toolchain's, so split functions link too.
+  toolchain's, so split functions link too.  The assembler preprocesses
+  like the toolchain (`#define`, `#include` with `-I`, `#ifdef`), generates
+  `go_asm.h` from the package's Go files, and carries `PCALIGN`, the
+  `LOCK`/`REP` prefixes and the literal-data pseudo-ops.
 - **Disassembler.**  `gasm dis` lists a `.s` file's functions at their real
   offsets after assembling, or disassembles raw bytes from a file or stdin.
 - **Dynamic verification.**  `gasm verify` JIT-loads assembled functions into
@@ -91,13 +94,16 @@ to give that syntax the tooling it deserves.
 - **Debugger.**  `gasm debug` is a source-level ptrace debugger with
   breakpoints (optionally conditional), hardware watchpoints, register and
   memory inspection, and headless script runs that report instruction and
-  label coverage.
+  label coverage; it runs on Linux (all four architectures) and FreeBSD
+  (amd64, arm64, riscv64).
 - **Language server.**  `gasm lsp` serves completion, hover, document symbols,
   push and pull diagnostics, semantic-token highlighting, go-to-definition,
   find references, rename, formatting, inlay hints, code actions, signature
   help, document highlights, workspace symbol search, #include document
   links and folding ranges over stdio; definition, references and rename
-  work across every open document.
+  work across every open document and the indexed workspace files beyond
+  them, and the quick fixes add a missing textflag.h include and set the
+  argument area from the // func signature.
 - **Comparators and audits.**  `gasm diff` compares the machine code of two
   assembly files byte-for-byte, `gasm profile` shows basic-block structure,
   `gasm audit-instructions` diffs the encoder against the installed toolchain,
@@ -110,9 +116,9 @@ Four architectures, the four that matter in practice:
 | Architecture | GOARCH      | File suffix  | Instructions recognised                     |
 |--------------|-------------|--------------|---------------------------------------------|
 | AMD64        | `amd64`     | `_amd64.s`   | 1600 + common opcodes + traditional aliases |
-| ARM64        | `arm64`     | `_arm64.s`   | 538 + common opcodes                        |
-| RISC-V       | `riscv64`   | `_riscv64.s` | 961 + common opcodes                        |
-| LoongArch    | `loong64`   | `_loong64.s` | 799 + common opcodes                        |
+| ARM64        | `arm64`     | `_arm64.s`   | 645 + common opcodes                        |
+| RISC-V       | `riscv64`   | `_riscv64.s` | 992 + common opcodes                        |
+| LoongArch    | `loong64`   | `_loong64.s` | 808 + common opcodes                        |
 
 "Common opcodes" are the instructions shared by every architecture (`RET`,
 `JMP`, `NOP`, `CALL`, `TEXT`, `FUNCDATA`, `PCDATA`, ...).  AMD64 additionally
@@ -124,10 +130,13 @@ can emit today is narrower, and a recognised but unencodable instruction is
 reported as an explicit error, never as a wrong byte.
 
 The same measurement runs over GOROOT's whole assembly corpus:
-`gasm audit-instructions --corpus` reports 136 of 433 attemptable files
-(31.4 %) assembling for every target architecture today (files named for
-other Go ports are counted but never attempted), with the top failure
-reasons per architecture; the number moves with every release.
+`gasm audit-instructions --corpus` reports every real-code GOROOT assembly
+file (the tree without testdata) assembling for every target its build
+admits: 250 of 250, 100 %. Over the whole tree including testdata the
+measure is 271 of 322 attemptable (84.2 %); files named for other Go ports
+are counted but never attempted, and `//go:build` constraints decide which
+targets attempt a file at all, exactly as the build does. The number moves
+with every release.
 
 ### Validation status
 
@@ -142,7 +151,7 @@ actually been executed.
 |---|---|---|
 | Encoding: byte-for-byte against `go tool asm` | native hardware | native hardware (the toolchain cross-assembles any GOARCH on any host) |
 | Execution: JIT calls, ABI checks, differential fuzzing | native hardware | qemu-user emulation |
-| Debugger: ptrace tracing, breakpoints, watchpoints, coverage | native hardware | emulation cannot run ptrace; the layer compiles and its architecture-neutral units run under `go test ./...`, nothing more |
+| Debugger: ptrace tracing, breakpoints, watchpoints, coverage | native hardware | emulation cannot run ptrace; the layer compiles and its architecture-neutral units run under `go test ./...`, nothing more. FreeBSD (amd64, arm64, riscv64) is in the same position: the port compiles behind the cross-build gate and its integration test is ready, but no FreeBSD machine has executed it |
 
 Consequences, stated plainly. An emulator is a model of a CPU, not the
 CPU: instruction semantics are implemented in software and can differ
@@ -186,8 +195,9 @@ verified the way the code is verified: an encoding documented here is
 one that differential tests against `go tool asm` confirm
 byte-for-byte, and a format field documented here is one the linker
 demonstrably reads. The work has begun: [docs/GOOBJ.md](docs/GOOBJ.md)
-specifies the object file format completely. The assembler language
-reference follows.
+specifies the object file format completely, and
+[docs/asm/README.md](docs/asm/README.md) opens the language reference
+with its common core. The per-architecture pages follow.
 
 ## Direction
 
@@ -211,9 +221,11 @@ The plan, in the order it is being worked:
   toolchain itself does not support; through ELF, Plan 9 assembly becomes
   usable outside Go entirely.
 - **Platforms: Linux and FreeBSD.**  Linux is supported today on all four
-  architectures and is where the binary builds.  FreeBSD follows: the
-  JIT's executable-memory mapping and the ptrace debugger layer are the
-  two pieces of porting work.  Other unix systems may follow those two.
+  architectures and is where the binary builds.  FreeBSD follows on amd64,
+  arm64 and riscv64: the JIT's executable-memory mapping and the ptrace
+  debugger layer are ported (the debugger's live validation awaits a
+  FreeBSD machine, as the validation status states).  Other unix systems
+  may follow those two.
 - **Four architectures, no more.**  amd64, arm64, riscv64 and loong64.
   No others are planned.
 
@@ -221,11 +233,11 @@ The plan, in the order it is being worked:
 
 Prebuilt binaries for linux/amd64, linux/arm64, linux/riscv64 and
 linux/loong64 are on the
-[releases page](https://sourcedock.dev/petrbalvin/gasm-devkit/releases).
+[releases page](https://sourcedock.dev/petrbalvin/gasm-sdk/releases).
 From source (Go 1.27.1):
 
 ```sh
-go install sourcedock.dev/petrbalvin/gasm-devkit/cmd/gasm@latest
+go install sourcedock.dev/petrbalvin/gasm-sdk/cmd/gasm@latest
 ```
 
 Or from a repository checkout:
@@ -315,6 +327,7 @@ recipe.
   them
 - [docs/ARCHITECTURE.md](docs/ARCHITECTURE.md): components and data flow
 - [docs/GOOBJ.md](docs/GOOBJ.md): the GOOBJ object file format specification
+- [docs/asm/](docs/asm/README.md): the Plan 9 assembly language reference
 - [docs/DEVELOPMENT.md](docs/DEVELOPMENT.md): development setup and recipes
 - [CHANGELOG.md](CHANGELOG.md): release history
 

SECURITY.md

@@ -7,7 +7,7 @@ releases do not receive them.
 
 | Version | Supported |
 |---|---|
-| 0.34.0 | yes |
+| 0.35.0 | yes |
 | older releases | no |
 
 ## Reporting a vulnerability

_gen/gen.go

@@ -5,9 +5,13 @@
 // toolchain's own assembler source.  Go's Plan 9 assembler defines the exact,
 // complete set of mnemonics it accepts for each architecture in
 // $GOROOT/src/cmd/internal/obj/<arch>/anames.go; this tool extracts those
-// names so gasm-devkit supports every instruction the real assembler does,
+// names so gasm-sdk supports every instruction the real assembler does,
 // with no hand-maintained (and therefore inevitably incomplete) lists.
 //
+// The same data feeds the generated instruction appendices of the assembly
+// language reference, docs/asm/INSTRUCTIONS-<ARCH>.md, so that the reference
+// cannot drift from the tables it documents.
+//
 // Usage (via the justfile):
 //
 //	just gen
@@ -26,9 +30,12 @@ import (
 	"path/filepath"
 	"sort"
 	"strings"
+
+	"sourcedock.dev/petrbalvin/gasm-sdk/arch"
+	"sourcedock.dev/petrbalvin/gasm-sdk/asm"
 )
 
-// archDirs maps a gasm-devkit architecture name to its obj sub-directory.
+// archDirs maps a gasm-sdk architecture name to its obj sub-directory.
 var archDirs = []struct {
 	arch string
 	sub  string
@@ -39,11 +46,30 @@ var archDirs = []struct {
 	{"loong64", "loong64"},
 }
 
+// docPages maps an architecture to its generated appendix in the language
+// reference.  The amd64 page carries a per-mnemonic encodability column,
+// decided by asm.Encodable, which mirrors the encoder's own dispatch; the
+// other targets have no single cheap predicate, so their pages carry the
+// inventory and point at the live measurement instead.
+var docPages = []struct {
+	arch      arch.Arch
+	title     string
+	file      string
+	anames    string
+	encodable bool
+}{
+	{arch.AMD64, "AMD64", "INSTRUCTIONS-AMD64.md", "cmd/internal/obj/x86/anames.go", true},
+	{arch.ARM64, "ARM64", "INSTRUCTIONS-ARM64.md", "cmd/internal/obj/arm64/anames.go", false},
+	{arch.RISCV, "RISC-V 64", "INSTRUCTIONS-RISCV64.md", "cmd/internal/obj/riscv/anames.go", false},
+	{arch.LOONG64, "LoongArch 64", "INSTRUCTIONS-LOONG64.md", "cmd/internal/obj/loong64/anames.go", false},
+}
+
 func main() {
 	goroot := strings.TrimSpace(runGoEnvGOROOT())
 	if goroot == "" {
 		fatal("could not determine GOROOT")
 	}
+	version := strings.TrimSpace(runGoEnv("GOVERSION"))
 	// The common opcodes shared by every architecture (RET, JMP, NOP, CALL,
 	// TEXT, FUNCDATA, …) live in cmd/internal/obj/util.go.
 	commonPath := filepath.Join(goroot, "src", "cmd", "internal", "obj", "util.go")
@@ -57,16 +83,24 @@ func main() {
 	}
 	fmt.Printf("%-8s %4d instructions -> arch/common_gen.go\n", "common", len(common))
 
+	names := map[string][]string{}
 	for _, a := range archDirs {
 		path := filepath.Join(goroot, "src", "cmd", "internal", "obj", a.sub, "anames.go")
-		names, err := extractInstrs(path)
+		names[a.arch], err = extractInstrs(path)
 		if err != nil {
 			fatal("extract %s: %v", a.arch, err)
 		}
-		if err := writeGen(a.arch, a.sub, names); err != nil {
+		if err := writeGen(a.arch, a.sub, names[a.arch]); err != nil {
 			fatal("write %s: %v", a.arch, err)
 		}
-		fmt.Printf("%-8s %4d instructions -> arch/%s_gen.go\n", a.arch, len(names), a.arch)
+		fmt.Printf("%-8s %4d instructions -> arch/%s_gen.go\n", a.arch, len(names[a.arch]), a.arch)
+	}
+
+	for _, p := range docPages {
+		if err := writeDocPage(p.arch, p.title, p.file, p.anames, version, p.encodable); err != nil {
+			fatal("write %s: %v", p.file, err)
+		}
+		fmt.Printf("%-8s -> docs/asm/%s\n", p.arch, p.file)
 	}
 }
 
@@ -85,7 +119,7 @@ func filterCommon(names []string) []string {
 // writeCommon emits arch/common_gen.go.
 func writeCommon(names []string) error {
 	var b strings.Builder
-	b.WriteString("// Code generated by gasm-devkit _gen; DO NOT EDIT.\n")
+	b.WriteString("// Code generated by gasm-sdk _gen; DO NOT EDIT.\n")
 	b.WriteString("// Source: cmd/internal/obj/util.go from the Go toolchain.\n")
 	b.WriteString("//\n")
 	b.WriteString("// Copyright (c) 2026 Petr Balvín <opensource@petrbalvin.org> (https://petrbalvin.org)\n")
@@ -156,7 +190,7 @@ func stringLit(elt ast.Expr) string {
 // writeGen emits arch/<arch>_gen.go.
 func writeGen(arch, sub string, names []string) error {
 	var b strings.Builder
-	b.WriteString("// Code generated by gasm-devkit _gen; DO NOT EDIT.\n")
+	b.WriteString("// Code generated by gasm-sdk _gen; DO NOT EDIT.\n")
 	b.WriteString("// Source: cmd/internal/obj/" + sub + "/anames.go from the Go toolchain.\n")
 	b.WriteString("//\n")
 	b.WriteString("// Copyright (c) 2026 Petr Balvín <opensource@petrbalvin.org> (https://petrbalvin.org)\n")
@@ -172,6 +206,61 @@ func writeGen(arch, sub string, names []string) error {
 	return os.WriteFile(filepath.Join("arch", arch+"_gen.go"), []byte(b.String()), 0o644)
 }
 
+// writeDocPage emits docs/asm/<file>, the generated instruction appendix of
+// the language reference for one architecture: every mnemonic the toolchain
+// accepts, with the curated summary where the architecture table carries one
+// and, on amd64, a per-mnemonic encodability column.
+func writeDocPage(a arch.Arch, title, file, anames, version string, encodable bool) error {
+	table := arch.ForArch(a)
+	instrs := table.Instructions()
+
+	var b strings.Builder
+	b.WriteString("# " + title + ": instruction inventory\n\n")
+	b.WriteString("Generated by gasm-sdk's `_gen` from the Go toolchain's instruction table\n")
+	b.WriteString("(`" + anames + "`, " + version + "); DO NOT EDIT. This page lists every mnemonic\n")
+	b.WriteString("`go tool asm` accepts on this target, which is the upper bound of the\n")
+	b.WriteString("language on it: a name absent here is not an instruction of the target,\n")
+	b.WriteString("and a name present here may still be one gasm's encoder cannot emit yet.\n\n")
+
+	encodableCount := 0
+	if encodable {
+		b.WriteString("The `gasm encodes` column reports whether gasm's encoder can emit the\n")
+		b.WriteString("mnemonic today; the gap is the encoder backlog, measured live by\n")
+		b.WriteString("`gasm audit-instructions`.\n\n")
+		b.WriteString("| Mnemonic | gasm encodes | Notes |\n")
+		b.WriteString("|---|---|---|\n")
+		for _, in := range instrs {
+			ok := asm.Encodable(in.Name)
+			if ok {
+				encodableCount++
+			}
+			b.WriteString("| `" + in.Name + "` | " + yesNo(ok) + " | " + in.Summary + " |\n")
+		}
+		b.WriteString("\n")
+		fmt.Fprintf(&b, "Recognised: %d mnemonics. gasm encodes: %d.\n", len(instrs), encodableCount)
+	} else {
+		b.WriteString("The inventory carries no per-mnemonic encoder column: on this target\n")
+		b.WriteString("encodability is decided per operand shape, and the live measured\n")
+		b.WriteString("coverage is reported by `gasm audit-instructions`.\n\n")
+		b.WriteString("| Mnemonic | Notes |\n")
+		b.WriteString("|---|---|\n")
+		for _, in := range instrs {
+			b.WriteString("| `" + in.Name + "` | " + in.Summary + " |\n")
+		}
+		b.WriteString("\n")
+		fmt.Fprintf(&b, "Recognised: %d mnemonics.\n", len(instrs))
+	}
+	return os.WriteFile(filepath.Join("docs", "asm", file), []byte(b.String()), 0o644)
+}
+
+// yesNo renders a boolean as the word the appendix tables use.
+func yesNo(v bool) string {
+	if v {
+		return "yes"
+	}
+	return "no"
+}
+
 func runGoEnvGOROOT() string {
 	out, err := exec.Command("go", "env", "GOROOT").Output()
 	if err != nil {
@@ -180,6 +269,15 @@ func runGoEnvGOROOT() string {
 	return string(out)
 }
 
+// runGoEnv runs `go env` for a single variable.
+func runGoEnv(name string) string {
+	out, err := exec.Command("go", "env", name).Output()
+	if err != nil {
+		return ""
+	}
+	return string(out)
+}
+
 func fatal(format string, args ...any) {
 	fmt.Fprintf(os.Stderr, "gen: "+format+"\n", args...)
 	os.Exit(1)

arch/amd64_gen.go

@@ -1,4 +1,4 @@
-// Code generated by gasm-devkit _gen; DO NOT EDIT.
+// Code generated by gasm-sdk _gen; DO NOT EDIT.
 // Source: cmd/internal/obj/x86/anames.go from the Go toolchain.
 //
 // Copyright (c) 2026 Petr Balvín <opensource@petrbalvin.org> (https://petrbalvin.org)

arch/arm64_gen.go

@@ -1,4 +1,4 @@
-// Code generated by gasm-devkit _gen; DO NOT EDIT.
+// Code generated by gasm-sdk _gen; DO NOT EDIT.
 // Source: cmd/internal/obj/arm64/anames.go from the Go toolchain.
 //
 // Copyright (c) 2026 Petr Balvín <opensource@petrbalvin.org> (https://petrbalvin.org)
@@ -364,6 +364,8 @@ var arm64GeneratedInstrs = []string{
 	"REVW",
 	"ROR",
 	"RORW",
+	"RPRFM",
+	"SB",
 	"SBC",
 	"SBCS",
 	"SBCSW",
@@ -477,23 +479,68 @@ var arm64GeneratedInstrs = []string{
 	"UXTH",
 	"UXTHW",
 	"UXTW",
+	"VABS",
 	"VADD",
 	"VADDP",
 	"VADDV",
 	"VAND",
 	"VBCAX",
+	"VBIC",
 	"VBIF",
 	"VBIT",
 	"VBSL",
+	"VCLS",
+	"VCLZ",
 	"VCMEQ",
+	"VCMGE",
+	"VCMGT",
+	"VCMHI",
+	"VCMHS",
+	"VCMLE",
+	"VCMLT",
 	"VCMTST",
 	"VCNT",
 	"VDUP",
 	"VEOR",
 	"VEOR3",
 	"VEXT",
+	"VFABS",
+	"VFADD",
+	"VFADDP",
+	"VFCMEQ",
+	"VFCMGE",
+	"VFCMGT",
+	"VFCMLE",
+	"VFCMLT",
+	"VFCVTL",
+	"VFCVTL2",
+	"VFCVTN",
+	"VFCVTN2",
+	"VFCVTZS",
+	"VFCVTZU",
+	"VFDIV",
+	"VFMAX",
+	"VFMAXNM",
+	"VFMAXNMP",
+	"VFMAXNMV",
+	"VFMAXP",
+	"VFMAXV",
+	"VFMIN",
+	"VFMINNM",
+	"VFMINNMP",
+	"VFMINNMV",
+	"VFMINP",
+	"VFMINV",
 	"VFMLA",
 	"VFMLS",
+	"VFMUL",
+	"VFNEG",
+	"VFRINTM",
+	"VFRINTN",
+	"VFRINTP",
+	"VFRINTZ",
+	"VFSQRT",
+	"VFSUB",
 	"VLD1",
 	"VLD1R",
 	"VLD2",
@@ -502,11 +549,17 @@ var arm64GeneratedInstrs = []string{
 	"VLD3R",
 	"VLD4",
 	"VLD4R",
+	"VMLA",
+	"VMLS",
 	"VMOV",
 	"VMOVD",
 	"VMOVI",
 	"VMOVQ",
 	"VMOVS",
+	"VMUL",
+	"VNEG",
+	"VNOT",
+	"VORN",
 	"VORR",
 	"VPMULL",
 	"VPMULL2",
@@ -515,14 +568,47 @@ var arm64GeneratedInstrs = []string{
 	"VREV16",
 	"VREV32",
 	"VREV64",
+	"VSCVTF",
+	"VSHADD",
 	"VSHL",
+	"VSHRN",
+	"VSHRN2",
 	"VSLI",
+	"VSMAX",
+	"VSMAXP",
+	"VSMAXV",
+	"VSMIN",
+	"VSMINP",
+	"VSMINV",
+	"VSMLAL",
+	"VSMLAL2",
+	"VSMLSL",
+	"VSMLSL2",
+	"VSMULL",
+	"VSMULL2",
+	"VSQABS",
+	"VSQADD",
+	"VSQNEG",
+	"VSQSHL",
+	"VSQSUB",
+	"VSQXTN",
+	"VSQXTN2",
+	"VSQXTUN",
+	"VSQXTUN2",
+	"VSRHADD",
 	"VSRI",
+	"VSRSHR",
+	"VSSHL",
+	"VSSHLL",
+	"VSSHLL2",
+	"VSSHR",
 	"VST1",
 	"VST2",
 	"VST3",
 	"VST4",
 	"VSUB",
+	"VSXTL",
+	"VSXTL2",
 	"VTBL",
 	"VTBX",
 	"VTRN1",
@@ -530,8 +616,27 @@ var arm64GeneratedInstrs = []string{
 	"VUADDLV",
 	"VUADDW",
 	"VUADDW2",
+	"VUCVTF",
+	"VUHADD",
 	"VUMAX",
+	"VUMAXP",
+	"VUMAXV",
 	"VUMIN",
+	"VUMINP",
+	"VUMINV",
+	"VUMLAL",
+	"VUMLAL2",
+	"VUMLSL",
+	"VUMLSL2",
+	"VUMULL",
+	"VUMULL2",
+	"VUQADD",
+	"VUQSHL",
+	"VUQSUB",
+	"VUQXTN",
+	"VUQXTN2",
+	"VURHADD",
+	"VUSHL",
 	"VUSHLL",
 	"VUSHLL2",
 	"VUSHR",
@@ -541,6 +646,8 @@ var arm64GeneratedInstrs = []string{
 	"VUZP1",
 	"VUZP2",
 	"VXAR",
+	"VXTN",
+	"VXTN2",
 	"VZIP1",
 	"VZIP2",
 	"WFE",

arch/common_gen.go

@@ -1,4 +1,4 @@
-// Code generated by gasm-devkit _gen; DO NOT EDIT.
+// Code generated by gasm-sdk _gen; DO NOT EDIT.
 // Source: cmd/internal/obj/util.go from the Go toolchain.
 //
 // Copyright (c) 2026 Petr Balvín <opensource@petrbalvin.org> (https://petrbalvin.org)

arch/loong64_gen.go

@@ -1,4 +1,4 @@
-// Code generated by gasm-devkit _gen; DO NOT EDIT.
+// Code generated by gasm-sdk _gen; DO NOT EDIT.
 // Source: cmd/internal/obj/loong64/anames.go from the Go toolchain.
 //
 // Copyright (c) 2026 Petr Balvín <opensource@petrbalvin.org> (https://petrbalvin.org)
@@ -152,6 +152,8 @@ var loong64GeneratedInstrs = []string{
 	"FNMADDF",
 	"FNMSUBD",
 	"FNMSUBF",
+	"FRINTD",
+	"FRINTF",
 	"FSCALEBD",
 	"FSCALEBF",
 	"FSEL",
@@ -177,7 +179,10 @@ var loong64GeneratedInstrs = []string{
 	"FTINTWF",
 	"JIRL",
 	"LL",
+	"LLACQV",
+	"LLACQW",
 	"LLV",
+	"LLW",
 	"LU12IW",
 	"LU32ID",
 	"LU52ID",
@@ -248,7 +253,11 @@ var loong64GeneratedInstrs = []string{
 	"ROTR",
 	"ROTRV",
 	"SC",
+	"SCQ",
+	"SCRELV",
+	"SCRELW",
 	"SCV",
+	"SCW",
 	"SGT",
 	"SGTU",
 	"SLL",

arch/riscv_gen.go

@@ -1,4 +1,4 @@
-// Code generated by gasm-devkit _gen; DO NOT EDIT.
+// Code generated by gasm-sdk _gen; DO NOT EDIT.
 // Source: cmd/internal/obj/riscv/anames.go from the Go toolchain.
 //
 // Copyright (c) 2026 Petr Balvín <opensource@petrbalvin.org> (https://petrbalvin.org)
@@ -81,6 +81,9 @@ var riscvGeneratedInstrs = []string{
 	"CLD",
 	"CLDSP",
 	"CLI",
+	"CLMUL",
+	"CLMULH",
+	"CLMULR",
 	"CLUI",
 	"CLW",
 	"CLWSP",
@@ -95,13 +98,20 @@ var riscvGeneratedInstrs = []string{
 	"CSDSP",
 	"CSLLI",
 	"CSRAI",
+	"CSRC",
+	"CSRCI",
 	"CSRLI",
+	"CSRR",
 	"CSRRC",
 	"CSRRCI",
 	"CSRRS",
 	"CSRRSI",
 	"CSRRW",
 	"CSRRWI",
+	"CSRS",
+	"CSRSI",
+	"CSRW",
+	"CSRWI",
 	"CSUB",
 	"CSUBW",
 	"CSW",
@@ -259,6 +269,7 @@ var riscvGeneratedInstrs = []string{
 	"ORCB",
 	"ORI",
 	"ORN",
+	"PAUSE",
 	"RDCYCLE",
 	"RDINSTRET",
 	"RDTIME",
@@ -322,6 +333,8 @@ var riscvGeneratedInstrs = []string{
 	"VADDVI",
 	"VADDVV",
 	"VADDVX",
+	"VANDNVV",
+	"VANDNVX",
 	"VANDVI",
 	"VANDVV",
 	"VANDVX",
@@ -329,8 +342,17 @@ var riscvGeneratedInstrs = []string{
 	"VASUBUVX",
 	"VASUBVV",
 	"VASUBVX",
+	"VBREV8V",
+	"VBREVV",
+	"VCLMULHVV",
+	"VCLMULHVX",
+	"VCLMULVV",
+	"VCLMULVX",
+	"VCLZV",
 	"VCOMPRESSVM",
 	"VCPOPM",
+	"VCPOPV",
+	"VCTZV",
 	"VDIVUVV",
 	"VDIVUVX",
 	"VDIVVV",
@@ -743,10 +765,16 @@ var riscvGeneratedInstrs = []string{
 	"VREMUVX",
 	"VREMVV",
 	"VREMVX",
+	"VREV8V",
 	"VRGATHEREI16VV",
 	"VRGATHERVI",
 	"VRGATHERVV",
 	"VRGATHERVX",
+	"VROLVV",
+	"VROLVX",
+	"VRORVI",
+	"VRORVV",
+	"VRORVX",
 	"VRSUBVI",
 	"VRSUBVX",
 	"VS1RV",
@@ -950,6 +978,9 @@ var riscvGeneratedInstrs = []string{
 	"VWMULVX",
 	"VWREDSUMUVS",
 	"VWREDSUMVS",
+	"VWSLLVI",
+	"VWSLLVV",
+	"VWSLLVX",
 	"VWSUBUVV",
 	"VWSUBUVX",
 	"VWSUBUWV",

asm/aarch64_goobj_test.go

@@ -11,7 +11,7 @@ import (
 	"strings"
 	"testing"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/parser"
+	"sourcedock.dev/petrbalvin/gasm-sdk/parser"
 )
 
 // TestGOObjectAARCH64Structure checks the basic structure of the emitted

asm/arm64_assemble.go

@@ -9,7 +9,7 @@ import (
 	"strconv"
 	"strings"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/ast"
+	"sourcedock.dev/petrbalvin/gasm-sdk/ast"
 )
 
 // assembleARM64 assembles an AArch64 (arm64) TEXT function body into machine
@@ -637,6 +637,20 @@ func encodeARM64Branch(mnem string, ops []*ast.Operand, pc int, offsets map[stri
 		return a64wordLE(a64UncondBranch(opc, uint32(rn), 0)), nil
 	}
 
+	// The bare spelling BL R9 is the same indirect branch: the parser reads
+	// a bare identifier as a symbol, and one named for a register is an
+	// indirect branch through it, which the toolchain accepts alongside the
+	// parenthesised form (BL (R3) and BL R3 both encode BLR R3).
+	if op.Addr.Sym != nil && op.Addr.Sym.Pseudo == "" && op.Addr.Base == "" && op.Addr.Index == "" {
+		if rn := arm64RegNum(op.Addr.Sym.Name); rn >= 0 {
+			opc := uint32(0) // BR
+			if link {
+				opc = 1 // BLR
+			}
+			return a64wordLE(a64UncondBranch(opc, uint32(rn), 0)), nil
+		}
+	}
+
 	// Symbol reference: BL sym(SB), or B sym(SB) for a tail call, against a
 	// relocation (R_CALLARM64 either way).
 	if op.Addr.Sym != nil && op.Addr.Sym.Pseudo == "SB" {
@@ -1454,6 +1468,15 @@ func encodeARM64Mov(instr *ast.Instr, mnem string, wb string, fi arm64FrameInfo,
 			}
 			return encodeARM64SBAddr(src.Imm.Sym, rd, relocs), nil
 		}
+		// Immediate → memory: only storing zero is encodable (the ZR
+		// register); the toolchain rejects any other immediate-to-memory
+		// combination ("illegal combination").
+		if isMemOperand(dst) {
+			if arm64Imm64(src) != 0 {
+				return nil, fmt.Errorf("%s: illegal combination: an immediate store must be zero", mnem)
+			}
+			return encodeARM64MemOp(mnem, dst, 31, false, fi, "")
+		}
 		rd := arm64RegNum(operandRegName(dst))
 		if rd < 0 {
 			return nil, fmt.Errorf("%s $imm: invalid destination register", mnem)
@@ -3046,29 +3069,29 @@ func encodeARM64MoveWide(mnem string, baseOp uint32, ops []*ast.Operand) ([]byte
 	// base, so MOVZ and MOVN come along for free.
 	opc := baseOp >> 29 & 3
 	sf := baseOp >> 31 & 1
-	v := arm64Imm64(ops[0])
-	if v < 0 {
-		return nil, fmt.Errorf("%s: negative immediate %d", mnem, v)
+	// The toolchain's optab case 33, shared by the whole family in both
+	// widths: the immediate is one unsigned 64-bit pattern (a high-lane
+	// constant such as $(40000<<48) arrives negative through int64
+	// folding), it must occupy exactly one 16-bit lane, zero is rejected,
+	// and the W forms cannot reach the top half.
+	u := uint64(arm64Imm64(ops[0]))
+	if u == 0 {
+		return nil, fmt.Errorf("%s: zero immediate cannot be handled", mnem)
 	}
 	hw := -1
-	for i := range 4 {
-		if v>>(uint(i)*16)&0xFFFF != 0 {
-			hw = i
+	for lane := range 4 {
+		if u&^(uint64(0xFFFF)<<(lane*16)) == 0 {
+			hw = lane
 			break
 		}
 	}
 	if hw < 0 {
-		hw = 0 // zero: every chunk is zero, hw = 0 carries it
-	}
-	for i := hw + 1; i < 4; i++ {
-		if v>>(uint(i)*16)&0xFFFF != 0 {
-			return nil, fmt.Errorf("%s: immediate %d does not fit one 16-bit chunk", mnem, v)
-		}
+		return nil, fmt.Errorf("%s: immediate %#x does not fit one 16-bit chunk", mnem, u)
 	}
 	if sf == 0 && hw > 1 {
-		return nil, fmt.Errorf("%s: immediate %d out of range for the 32-bit form", mnem, v)
+		return nil, fmt.Errorf("%s: immediate %#x out of range for the 32-bit form", mnem, u)
 	}
-	return a64wordLE(a64MoveWide(sf, opc, uint32(hw), uint32(v>>uint(hw*16)&0xFFFF), uint32(rd))), nil
+	return a64wordLE(a64MoveWide(sf, opc, uint32(hw), uint32(u>>uint(hw*16)&0xFFFF), uint32(rd))), nil
 }
 
 // ---- Bitfield/EXTR encoding ----

asm/arm64_encode.go

@@ -33,7 +33,7 @@ import (
 	"strconv"
 	"strings"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/ast"
+	"sourcedock.dev/petrbalvin/gasm-sdk/ast"
 )
 
 // arm64RegNum returns the 5-bit register number for an AArch64 register name:

asm/arm64_encode_test.go

@@ -7,8 +7,8 @@ import (
 	"strings"
 	"testing"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/ast"
-	"sourcedock.dev/petrbalvin/gasm-devkit/parser"
+	"sourcedock.dev/petrbalvin/gasm-sdk/ast"
+	"sourcedock.dev/petrbalvin/gasm-sdk/parser"
 )
 
 func TestArm64LDRSTREncoding(t *testing.T) {
@@ -1051,6 +1051,41 @@ func TestArm64MOVK(t *testing.T) {
 	}
 }
 
+// TestArm64MOVKHighLane pins the shifted high-lane immediate the arm64 test
+// kernels write: $(40000<<48) folds to a negative int64, and the toolchain
+// reads the value as an unsigned 64-bit pattern when it picks the lane.
+func TestArm64MOVKHighLane(t *testing.T) {
+	got := arm64Words(t, "\tMOVK $(40000<<48), R0\n\tMOVK $0x9c40000000000000, R1\n")
+	want := []uint32{
+		0xf2f38800, // MOVK $(40000<<48), R0 (go tool asm: f2f38800)
+		0xf2f38801, // MOVK hw=3
+		0xd65f03c0,
+	}
+	if len(got) != len(want) {
+		t.Fatalf("word count = %d, want %d", len(got), len(want))
+	}
+	for i := range want {
+		if got[i] != want[i] {
+			t.Errorf("word %d = %08x, want %08x", i, got[i], want[i])
+		}
+	}
+}
+
+// TestArm64MoveWideZeroImmediate pins the toolchain's rejection of a zero
+// immediate in the move-wide family (optab case 33: "zero shifts cannot be
+// handled"): every lane is zero, so no hw field can carry it.
+func TestArm64MoveWideZeroImmediate(t *testing.T) {
+	for _, mnem := range []string{"MOVK", "MOVZ", "MOVN"} {
+		f, errs := parser.Parse("test_arm64.s", "#include \"textflag.h\"\n\nTEXT ·f(SB), NOSPLIT, $0-0\n\t"+mnem+" $0, R0\n\tRET\n")
+		if len(errs) > 0 {
+			t.Fatalf("%s: parse: %v", mnem, errs)
+		}
+		if _, err := AssembleFileARM64(f); err == nil {
+			t.Errorf("%s $0: expected error, got nil", mnem)
+		}
+	}
+}
+
 // TestArm64LoadImm64 tests 64-bit immediate loading.
 func TestArm64LoadImm64(t *testing.T) {
 	src := `#include "textflag.h"

asm/arm64_frame.go

@@ -53,7 +53,7 @@ package asm
 import (
 	"strings"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/ast"
+	"sourcedock.dev/petrbalvin/gasm-sdk/ast"
 )
 
 // arm64FrameInfo holds the frame layout derived from a TEXT directive.

asm/arm64_reloc_test.go

@@ -7,7 +7,7 @@ import (
 	"encoding/binary"
 	"testing"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/parser"
+	"sourcedock.dev/petrbalvin/gasm-sdk/parser"
 )
 
 // parseArm64File is a helper assembling one arm64 source file.

asm/assemble.go

@@ -8,7 +8,7 @@ import (
 	"strconv"
 	"strings"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/ast"
+	"sourcedock.dev/petrbalvin/gasm-sdk/ast"
 )
 
 // Assemble encodes the body of a TEXT function into x86-64 machine code,
@@ -38,10 +38,25 @@ func Assemble(t *ast.Text) ([]byte, map[string]int, error) {
 // rejects SB operands outright (single-function assembly cannot resolve
 // them).  When allowExternal is set, a reference to a symbol no GLOBL in the
 // file defines is recorded as an external relocation instead of failing
-// the object-file emitters resolve it at link time.
+// the object-file emitters resolve it at link time.  goos selects the TLS
+// access form: the empty default behaves as linux.
 type linkInfo struct {
 	symbols       map[string]bool
 	allowExternal bool
+	goos          string
+}
+
+// tlsOneInsn reports the one-instruction TLS form, obj6.go's
+// CanUse1InsnTLS for the GOOS gasm supports: the bare TLS load nops out and
+// the (TLS*1) index folds to a segment-absolute access.  Windows and plan9
+// keep the two-instruction form; shared linux does too, which gasm's raw
+// path does not model and therefore does not select.
+func (l *linkInfo) tlsOneInsn() bool {
+	switch l.goos {
+	case "", "linux", "freebsd":
+		return true
+	}
+	return false
 }
 
 // sbPatch is a function-relative static-symbol relocation: the disp32 field
@@ -86,6 +101,10 @@ func assemble(t *ast.Text, link *linkInfo) ([]byte, []sbPatch, map[string]int, [
 	// outgrows the short form.
 	long := make([]bool, len(t.Body))
 	sizes := make([]int, len(t.Body))
+	numTargets := make([]int, len(t.Body))
+	for i := range numTargets {
+		numTargets[i] = -1
+	}
 	offsets := map[string]int{}
 	pcs := make([]int, len(t.Body))
 	var guardJBlong, guardJBElong, moreJMPlong bool
@@ -94,23 +113,106 @@ func assemble(t *ast.Text, link *linkInfo) ([]byte, []sbPatch, map[string]int, [
 	for {
 		guard := fi.guardLen(guardJBlong, guardJBElong)
 		pos := guard + len(fi.prologue)
+		for i := range numTargets {
+			numTargets[i] = -1
+		}
+		idxAtPc := map[int]int{}
 		for i, stmt := range t.Body {
 			switch s := stmt.(type) {
 			case *ast.Label:
 				offsets[s.Name.Text] = pos
 			case *ast.Instr:
+				if strings.ToUpper(s.Mnemonic.Text) == "PCALIGN" {
+					// The alignment pseudo-statement: its size is the
+					// padding to the next boundary at this very position,
+					// filled with NOPs at emission.
+					pad, err := pcAlignPad(pcAlignValue(s), pos)
+					if err != nil {
+						return nil, nil, nil, nil, nil, nil, fmt.Errorf("PCALIGN: %w", err)
+					}
+					sizes[i] = pad
+					pcs[i] = pos
+					pos += pad
+					continue
+				}
 				sz, err := instrSize(s, fi, long[i], link)
 				if err != nil {
 					return nil, nil, nil, nil, nil, nil, fmt.Errorf("%s: %w", s.Mnemonic.Text, err)
 				}
 				sizes[i] = sz
 				pcs[i] = pos
+				idxAtPc[pos] = i
 				pos += sz
 			}
 		}
 		bodyLen := pos - (guard + len(fi.prologue))
 		// Expand any short jump whose displacement no longer fits rel8.
 		changed := false
+		// Numeric ±N(PC) jumps resolve against this iteration's layout; the
+		// emission pass reads the same table after the loop converges.  A
+		// target that is itself an unconditional local JMP is chased to the
+		// ultimate target: the toolchain's brloop pass collapses branch-to-
+		// branch chains before it encodes, so matching its bytes requires
+		// the same redirection.
+		for i := range numTargets {
+			numTargets[i] = -1
+		}
+		for i, stmt := range t.Body {
+			s, ok := stmt.(*ast.Instr)
+			if !ok {
+				continue
+			}
+			if len(s.Operands) == 1 {
+				if n, isNum := pcJumpOffset(s.Operands[0]); isNum {
+					if target, okT := pcJumpTarget(t, i, n, pcs); okT {
+						numTargets[i] = target
+					}
+				}
+			}
+		}
+		for i := range numTargets {
+			if numTargets[i] < 0 {
+				continue
+			}
+			tgt := numTargets[i]
+			for hop := 0; hop < len(t.Body); hop++ {
+				idx, ok := idxAtPc[tgt]
+				if !ok {
+					break
+				}
+				in, ok := t.Body[idx].(*ast.Instr)
+				if !ok || strings.ToUpper(in.Mnemonic.Text) != "JMP" || len(in.Operands) != 1 {
+					break
+				}
+				if name, isLabel := labelName(in.Operands[0]); isLabel {
+					tgt = offsets[resolve(name)]
+					continue
+				}
+				if n, isNum := pcJumpOffset(in.Operands[0]); isNum {
+					next, okT := pcJumpTarget(t, idx, n, pcs)
+					if !okT {
+						break
+					}
+					tgt = next
+					continue
+				}
+				break // JMP through a register or memory: the chain ends
+			}
+			numTargets[i] = tgt
+		}
+		for i, stmt := range t.Body {
+			s, ok := stmt.(*ast.Instr)
+			if !ok {
+				continue
+			}
+			if numTargets[i] >= 0 && !long[i] {
+				rel := int64(numTargets[i] - (pcs[i] + jumpSize(strings.ToUpper(s.Mnemonic.Text), false)))
+				if !fits8(rel) {
+					long[i] = true
+					changed = true
+				}
+			}
+		}
 		for i, stmt := range t.Body {
 			s, ok := stmt.(*ast.Instr)
 			if !ok {
@@ -232,7 +334,7 @@ func assemble(t *ast.Text, link *linkInfo) ([]byte, []sbPatch, map[string]int, [
 				spadjStep{pos + epi, 0},
 			)
 		}
-		code, ps, pool, err := encodeInstr(s, pos, offsets, fi, long[i], resolve, link)
+		code, ps, pool, err := encodeInstr(s, pos, offsets, fi, long[i], resolve, link, numTargets[i])
 		if err != nil {
 			return nil, nil, nil, nil, nil, nil, fmt.Errorf("%s: %w", s.Mnemonic.Text, err)
 		}
@@ -428,6 +530,99 @@ func computeFrame(t *ast.Text) frameInfo {
 	return fi
 }
 
+// pcJumpOffset recognises the numeric relative jump operand ±N(PC) and
+// returns N: the toolchain counts instructions, not bytes, so +2(PC) targets
+// the second instruction boundary after the branch.
+func pcJumpOffset(op *ast.Operand) (int, bool) {
+	if op.Kind != ast.OpAddr || op.Addr.Base != "PC" {
+		return 0, false
+	}
+	return int(op.Addr.Offset), true
+}
+
+// pcJumpTarget resolves a numeric jump at statement index j: N counts the
+// instruction statements after the jump itself (N = 0 is the jump's own
+// address, the classic park loop), and the target is the start of the Nth
+// one.  A negative N counts the same way backwards, before the jump: the
+// exit loops write JMP -3(PC) to land three instructions earlier.  Labels
+// count not, in either direction.  It reports false when the count runs
+// past the end of the function, or before its first instruction.
+func pcJumpTarget(t *ast.Text, j, n int, pcs []int) (int, bool) {
+	if n == 0 {
+		return pcs[j], true
+	}
+	if n < 0 {
+		seen := 0
+		for k := j - 1; k >= 0; k-- {
+			if _, ok := t.Body[k].(*ast.Instr); !ok {
+				continue
+			}
+			seen--
+			if seen == n {
+				return pcs[k], true
+			}
+		}
+		return 0, false
+	}
+	seen := 0
+	for k := j + 1; k < len(t.Body); k++ {
+		if _, ok := t.Body[k].(*ast.Instr); !ok {
+			continue
+		}
+		seen++
+		if seen == n {
+			return pcs[k], true
+		}
+	}
+	return 0, false
+}
+
+// x86 NOP encodings, single-instruction no-ops of lengths 1 to 9 (the
+// toolchain's asm6.go nop table); longer padding repeats the largest that
+// fits, greedy from the end.
+var x86Nops = [][]byte{
+	{0x90},
+	{0x66, 0x90},
+	{0x0F, 0x1F, 0x00},
+	{0x0F, 0x1F, 0x40, 0x00},
+	{0x0F, 0x1F, 0x44, 0x00, 0x00},
+	{0x66, 0x0F, 0x1F, 0x44, 0x00, 0x00},
+	{0x0F, 0x1F, 0x80, 0x00, 0x00, 0x00, 0x00},
+	{0x0F, 0x1F, 0x84, 0x00, 0x00, 0x00, 0x00, 0x00},
+	{0x66, 0x0F, 0x1F, 0x84, 0x00, 0x00, 0x00, 0x00, 0x00},
+}
+
+// fillNOPs fills p with the greedy largest single-instruction NOPs, exactly
+// the toolchain's fillnop.
+func fillNOPs(p []byte) {
+	for len(p) > 0 {
+		m := min(len(p), len(x86Nops))
+		copy(p[:m], x86Nops[m-1])
+		p = p[m:]
+	}
+}
+
+// pcAlignPad computes the padding PCALIGN $align inserts at pos: the
+// alignment must be a power of two in [8, 2048] and the padding runs to the
+// next boundary (zero when the position is already aligned).
+func pcAlignPad(align, pos int) (int, error) {
+	if align <= 0 || align&(align-1) != 0 || align < 8 || align > 2048 {
+		return 0, fmt.Errorf("alignment value of an instruction must be a power of two and in the range [8, 2048], got %d", align)
+	}
+	if lob := pos & (align - 1); lob != 0 {
+		return align - lob, nil
+	}
+	return 0, nil
+}
+
+// pcAlignValue reads a PCALIGN statement's alignment operand.
+func pcAlignValue(s *ast.Instr) int {
+	if len(s.Operands) == 1 && s.Operands[0].Kind == ast.OpImmediate && s.Operands[0].Imm.HasVal {
+		return int(s.Operands[0].Imm.Val)
+	}
+	return 0 // rejected by pcAlignPad's range check
+}
+
 // hasCall reports whether the function body contains a CALL instruction.
 func hasCall(t *ast.Text) bool {
 	for _, stmt := range t.Body {
@@ -602,7 +797,7 @@ func instrSize(s *ast.Instr, fi frameInfo, long bool, link *linkInfo) (int, erro
 		}
 		return jumpSize(mnem, long), nil
 	}
-	code, _, _, err := encodeInstr(s, 0, nil, fi, false, nil, link)
+	code, _, _, err := encodeInstr(s, 0, nil, fi, false, nil, link, -1)
 	if err != nil {
 		return 0, err
 	}
@@ -637,9 +832,21 @@ func jumpSize(mnem string, long bool) int {
 // (relative to pc, the instruction's own offset).  A RET in a frame-pointer
 // function is prefixed with the epilogue.  resolve, when non-nil, redirects a
 // jump label through the jump-to-jump chain before the offset lookup.
-func encodeInstr(s *ast.Instr, pc int, offsets map[string]int, fi frameInfo, long bool, resolve func(string) string, link *linkInfo) ([]byte, []sbPatch, []floatPoolEntry, error) {
+func encodeInstr(s *ast.Instr, pc int, offsets map[string]int, fi frameInfo, long bool, resolve func(string) string, link *linkInfo, numTarget int) ([]byte, []sbPatch, []floatPoolEntry, error) {
 	mnem := strings.ToUpper(s.Mnemonic.Text)
 
+	if mnem == "PCALIGN" {
+		// The layout pass already accounted the padding; emit the same
+		// amount of NOP bytes for the statement's own position.
+		pad, err := pcAlignPad(pcAlignValue(s), pc)
+		if err != nil {
+			return nil, nil, nil, err
+		}
+		out := make([]byte, pad)
+		fillNOPs(out)
+		return out, nil, nil, nil
+	}
+
 	var prefix []byte
 	if mnem == "RET" && fi.useFP {
 		prefix = fi.epilogue
@@ -677,7 +884,7 @@ func encodeInstr(s *ast.Instr, pc int, offsets map[string]int, fi frameInfo, lon
 			}
 			return append(prefix, code...), nil, nil, nil
 		}
-		code, err = encodeJump(s, mnem, pc+len(prefix), offsets, long, resolve)
+		code, err = encodeJump(s, mnem, pc+len(prefix), offsets, long, resolve, numTarget)
 	} else {
 		code, ps, pool, err = encodeNormal(s, fi, link)
 	}
@@ -703,6 +910,41 @@ func encodeNormal(s *ast.Instr, fi frameInfo, link *linkInfo) ([]byte, []sbPatch
 		}
 		return code, nil, nil, nil
 	}
+	// MOVQ $sym±off(SB), r64: the toolchain assembles a symbol immediate as
+	// LEAQ disp32(RIP), r64 with an R_PCREL relocation at the disp32 field,
+	// never as a 64-bit absolute immediate (verified against go tool asm).
+	// MOVD is the MOVQ alias; the narrower widths reject the form outright.
+	if (mnemUpper == "MOVQ" || mnemUpper == "MOVD") && len(s.Operands) == 2 &&
+		s.Operands[0].Kind == ast.OpImmediate && s.Operands[0].Imm.Sym != nil &&
+		s.Operands[0].Imm.Sym.Pseudo == "SB" {
+		mem := &ast.Operand{Kind: ast.OpAddr, Addr: ast.Address{Sym: s.Operands[0].Imm.Sym}}
+		src, err := operandFromAST(mnemUpper, mem, 8, fi, link)
+		if err != nil {
+			return nil, nil, nil, err
+		}
+		dst, err := operandFromAST(mnemUpper, s.Operands[1], 8, fi, link)
+		if err != nil {
+			return nil, nil, nil, err
+		}
+		e := &enc{}
+		if err := e.encodeLea([]Operand{src, dst}, 8); err != nil {
+			return nil, nil, nil, err
+		}
+		ps := make([]sbPatch, len(e.patches))
+		for i, p := range e.patches {
+			ps[i] = sbPatch{off: p.off, name: p.name, addend: p.addend}
+		}
+		return e.out, ps, nil, nil
+	}
+	// MOVQ/MOVL TLS, r: the bare TLS load.  The toolchain's progedit nops
+	// it out on the one-instruction TLS systems (linux and freebsd, not
+	// shared) and encodes the segment-prefixed load elsewhere; get_tls(r),
+	// the macro GOROOT's go_tls.h defines, expands to exactly this
+	// statement, and the toolchain's pairing pass removes it whenever the
+	// following instruction's (TLS*1) index folds.
+	if (mnemUpper == "MOVQ" || mnemUpper == "MOVL") && len(s.Operands) == 2 && isBareTLS(s.Operands[0]) {
+		return encodeTLSBaseLoad(s, fi, link)
+	}
 	_, size := splitSize(mnemUpper)
 	if size == 0 {
 		size = 8
@@ -722,10 +964,67 @@ func encodeNormal(s *ast.Instr, fi frameInfo, link *linkInfo) ([]byte, []sbPatch
 	ps := make([]sbPatch, len(e.patches))
 	for i, p := range e.patches {
 		ps[i] = sbPatch{off: p.off, name: p.name, addend: p.addend}
+		if p.tls {
+			ps[i].kind = RelTLSLE
+		}
 	}
 	return e.out, ps, e.floatPoolList(), nil
 }
 
+// isBareTLS reports whether the operand is the bare TLS pseudo-register
+// load source, the expansion of go_tls.h's get_tls(r) macro.
+func isBareTLS(op *ast.Operand) bool {
+	return op.Kind == ast.OpAddr && op.Addr.Sym != nil &&
+		op.Addr.Sym.Pseudo == "" && op.Addr.Sym.Name == "TLS" &&
+		op.Addr.Base == "" && op.Addr.Index == ""
+}
+
+// encodeTLSBaseLoad assembles MOVQ/MOVL TLS, r.  On the one-instruction TLS
+// systems (linux and freebsd outside -shared, obj6.go's CanUse1InsnTLS) the
+// statement nops out: the following (TLS*1) access folds to a direct
+// segment-absolute load.  The two-instruction systems keep the segment load,
+// nine bytes with the R_TLSLE patch site at the disp32.
+func encodeTLSBaseLoad(s *ast.Instr, fi frameInfo, link *linkInfo) ([]byte, []sbPatch, []floatPoolEntry, error) {
+	_, size := splitSize(strings.ToUpper(s.Mnemonic.Text))
+	if size == 0 {
+		size = 8
+	}
+	dst, err := operandFromAST("MOVQ", s.Operands[1], 8, fi, link)
+	if err != nil {
+		return nil, nil, nil, err
+	}
+	reg, ok := dst.(Reg)
+	if !ok || reg.isVec() {
+		return nil, nil, nil, fmt.Errorf("TLS: destination must be a general register")
+	}
+	if link == nil || link.tlsOneInsn() {
+		return nil, nil, nil, nil // noped out
+	}
+	seg := byte(0x64) // FS
+	if link.goos == "windows" {
+		seg = 0x65 // GS
+	}
+	e := &enc{}
+	i := &instr{
+		prefix: seg,
+		rexW:   size == 8,
+		rexR:   reg.idx >= 8,
+		opcode: []byte{0x8B},
+		modrm:  0x04 | (reg.idx&7)<<3,
+		sib:    0x25,
+		disp:   le32(0),
+		tls:    true,
+	}
+	if err := e.emit(i); err != nil {
+		return nil, nil, nil, err
+	}
+	ps := make([]sbPatch, len(e.patches))
+	for i, p := range e.patches {
+		ps[i] = sbPatch{off: p.off, name: p.name, addend: p.addend, kind: RelTLSLE}
+	}
+	return e.out, ps, nil, nil
+}
+
 // encodeBookkeeping accepts-and-ignores FUNCDATA and PCDATA at the statement
 // level, before operand conversion: the toolchain's shapes are FUNCDATA
 // $n, sym(SB) and PCDATA $n, $m, and neither contributes a byte to the
@@ -753,22 +1052,30 @@ func encodeBookkeeping(upper string, s *ast.Instr) ([]byte, error) {
 }
 
 // encodeJump encodes a JMP/CALL/Jcc with a relative offset resolved from the
-// target label, in the short (rel8) or long (rel32) form.
-func encodeJump(s *ast.Instr, mnem string, pc int, offsets map[string]int, long bool, resolve func(string) string) ([]byte, error) {
+// target label or from a numeric ±N(PC) instruction count, in the short
+// (rel8) or long (rel32) form.  numTarget is the resolved byte offset of a
+// numeric operand, negative when the operand is not one.
+func encodeJump(s *ast.Instr, mnem string, pc int, offsets map[string]int, long bool, resolve func(string) string, numTarget int) ([]byte, error) {
 	if len(s.Operands) != 1 {
 		return nil, fmt.Errorf("jump expects 1 operand, got %d", len(s.Operands))
 	}
-	name, ok := labelName(s.Operands[0])
-	if !ok {
+	name, isLabel := labelName(s.Operands[0])
+	if !isLabel && numTarget < 0 {
 		return nil, fmt.Errorf("jump target must be a local label")
 	}
+	var target int
+	if isLabel {
 		if resolve != nil && mnem != "CALL" {
 			name = resolve(name)
 		}
-	target, ok := offsets[name]
+		t, ok := offsets[name]
 		if !ok {
 			return nil, fmt.Errorf("undefined label %q", name)
 		}
+		target = t
+	} else {
+		target = numTarget
+	}
 	rel := int64(target - (pc + jumpSize(mnem, long)))
 
 	if !long {
@@ -841,6 +1148,11 @@ func indirectJumpTarget(s *ast.Instr) bool {
 		return false
 	}
 	a := s.Operands[0].Addr
+	// ±N(PC) is the numeric relative form, the PC counts instructions from
+	// the branch: relative, not indirect.
+	if a.Base == "PC" || a.Index == "PC" {
+		return false
+	}
 	if a.Base != "" || a.Index != "" {
 		return true
 	}
@@ -958,12 +1270,44 @@ func operandFromAST(mnemUpper string, op *ast.Operand, size int, fi frameInfo, l
 
 		// Memory with a real base register: (base), off(base), (base)(index*scale).
 		if a.Base != "" {
+			// Segment-absolute: 0x30(GS) and 0x28(FS), the windows TLS
+			// spellings.  The segment override prefixes a disp32 absolute
+			// reference with no relocation.
+			if a.Base == "GS" || a.Base == "FS" {
+				seg := byte(0x64)
+				if a.Base == "GS" {
+					seg = 0x65
+				}
+				return SegAbs{Disp: a.Offset, Size: size, Seg: seg}, nil
+			}
 			base, ok := ParseReg(a.Base)
 			if !ok {
 				return nil, fmt.Errorf("unknown base register %q", a.Base)
 			}
 			m := Mem{Base: base, Disp: a.Offset, HasBase: true, Size: size}
 			if a.Index != "" {
+				if a.Index == "TLS" {
+					// off(base)(TLS*1): the thread-local annotation.  The
+					// one-instruction TLS form folds it to off(TLS), the
+					// segment-prefixed absolute whose disp32 carries an
+					// R_TLS_LE patch site; the base register disappears
+					// from the encoding, exactly as the toolchain's
+					// progedit rewrites the address.
+					seg := byte(0x64) // FS on linux, freebsd, plan9
+					if link != nil && link.goos == "windows" {
+						seg = 0x65 // GS
+					}
+					return TLSMem{Disp: a.Offset, Size: size, Seg: seg}, nil
+				}
+				if a.Index == "GS" || a.Index == "FS" {
+					// 0(CX)(GS): the segment annotation rides the base
+					// access as the override prefix.
+					m.Seg = 0x64
+					if a.Index == "GS" {
+						m.Seg = 0x65
+					}
+					return m, nil
+				}
 				idx, ok := ParseReg(a.Index)
 				if !ok {
 					return nil, fmt.Errorf("unknown index register %q", a.Index)
@@ -984,6 +1328,11 @@ func operandFromAST(mnemUpper string, op *ast.Operand, size int, fi frameInfo, l
 			}
 			return Mem{Index: idx, Scale: a.Scale, Disp: a.Offset, HasIndex: true, Size: size}, nil
 		}
+		// A bare displacement with no base: the absolute address form,
+		// MOVL $0xf1, 0xf1.  No segment and no relocation.
+		if a.Sym == nil && a.Base == "" && a.Index == "" && a.HasOff {
+			return SegAbs{Disp: a.Offset, Size: size}, nil
+		}
 		// Bare register.
 		if a.Sym != nil && a.Sym.Pseudo == "" && a.Sym.Name != "" {
 			if r, ok := ParseReg(a.Sym.Name); ok {

asm/assemble_test.go

@@ -10,8 +10,8 @@ import (
 
 	"golang.org/x/arch/x86/x86asm"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/ast"
-	"sourcedock.dev/petrbalvin/gasm-devkit/parser"
+	"sourcedock.dev/petrbalvin/gasm-sdk/ast"
+	"sourcedock.dev/petrbalvin/gasm-sdk/parser"
 )
 
 // firstText parses src and returns its first TEXT function.
@@ -370,6 +370,58 @@ end:
 	}
 }
 
+// TestAssembleNumericPCJumps pins the numeric ±N(PC) branch operands: N
+// counts instruction statements, skipping labels, in both directions (the
+// runtime's exit loops write JMP -3(PC)), N = 0 parks on the jump itself.
+func TestAssembleNumericPCJumps(t *testing.T) {
+	fn := firstText(t, `
+#include "textflag.h"
+TEXT ·exit(SB), NOSPLIT, $0
+	MOVB $1, AL
+lab:
+	MOVB $2, AL
+	MOVB $3, AL
+	JMP -3(PC)
+	MOVB $4, AL
+park:
+	JMP 0(PC)
+	MOVB $5, AL
+	JMP 2(PC)
+	MOVB $6, AL
+	RET
+`)
+	code, _, err := Assemble(fn)
+	if err != nil {
+		t.Fatalf("Assemble: %v", err)
+	}
+	// From the Go-assembled function:
+	//   MOVB $1, AL       b001
+	//   MOVB $2, AL       b002
+	//   MOVB $3, AL       b003
+	//   JMP -3(PC)        ebf8      (three instructions back, past lab:)
+	//   MOVB $4, AL       b004
+	//   JMP 0(PC)         ebfe      (the park loop)
+	//   MOVB $5, AL       b005
+	//   JMP 2(PC)         eb02      (over MOVB $6 to the RET)
+	//   MOVB $6, AL       b006
+	//   RET               c3
+	want := []byte{
+		0xb0, 0x01,
+		0xb0, 0x02,
+		0xb0, 0x03,
+		0xeb, 0xf8,
+		0xb0, 0x04,
+		0xeb, 0xfe,
+		0xb0, 0x05,
+		0xeb, 0x02,
+		0xb0, 0x06,
+		0xc3,
+	}
+	if hexBytes(code) != hexBytes(want) {
+		t.Errorf("numeric-PC mismatch:\n got:  %s\n want: %s", hexBytes(code), hexBytes(want))
+	}
+}
+
 func TestAssemblePrefetch(t *testing.T) {
 	fn := firstText(t, `
 #include "textflag.h"

asm/elf_dwarf_test.go

@@ -8,7 +8,7 @@ import (
 	"encoding/binary"
 	"testing"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/parser"
+	"sourcedock.dev/petrbalvin/gasm-sdk/parser"
 )
 
 // ulebIter reads ULEB128 values, the .debug_abbrev and line-header

asm/elf_test.go

@@ -12,8 +12,8 @@ import (
 	"path/filepath"
 	"testing"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/ast"
-	"sourcedock.dev/petrbalvin/gasm-devkit/parser"
+	"sourcedock.dev/petrbalvin/gasm-sdk/ast"
+	"sourcedock.dev/petrbalvin/gasm-sdk/parser"
 )
 
 // The object-file tests share one source: two exported functions, one

asm/elfarm64_test.go

@@ -9,7 +9,7 @@ import (
 	"encoding/binary"
 	"testing"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/parser"
+	"sourcedock.dev/petrbalvin/gasm-sdk/parser"
 )
 
 // TestELFAARCH64Object checks the structure of the emitted AArch64 ELF64

asm/elfloong64_test.go

@@ -9,7 +9,7 @@ import (
 	"encoding/binary"
 	"testing"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/parser"
+	"sourcedock.dev/petrbalvin/gasm-sdk/parser"
 )
 
 // TestELFLOONG64Object checks the structure of the emitted LoongArch ELF64

asm/elfriscv_test.go

@@ -9,7 +9,7 @@ import (
 	"encoding/binary"
 	"testing"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/parser"
+	"sourcedock.dev/petrbalvin/gasm-sdk/parser"
 )
 
 // TestELFRISCVObjectDataRelocation checks that a symbol-valued DATA field

asm/encode.go

@@ -64,6 +64,7 @@ type encPatch struct {
 	off    int
 	name   string
 	addend int64
+	tls    bool // a TLS slot offset: the patch is R_TLSLE with no symbol
 }
 
 func (e *enc) encode(mnem string, ops []Operand) error {
@@ -578,6 +579,7 @@ type instr struct {
 	disp      []byte
 	imm       []byte
 	sb        *sbRef // static-symbol displacement in disp, awaiting resolution
+	tls       bool   // the displacement is a TLS slot offset, patched R_TLSLE
 }
 
 // sbRef records that an instruction's displacement refers to a static symbol
@@ -620,6 +622,9 @@ func (e *enc) emit(i *instr) error {
 	if i.sb != nil {
 		e.patches = append(e.patches, encPatch{off: len(e.out), name: i.sb.name, addend: i.sb.addend})
 	}
+	if i.tls {
+		e.patches = append(e.patches, encPatch{off: len(e.out), tls: true})
+	}
 	e.out = append(e.out, i.disp...)
 	e.out = append(e.out, i.imm...)
 	return nil
@@ -674,12 +679,30 @@ func setRMReg(i *instr, regField int, rexR, regForced bool, rm Operand, opSize i
 		i.disp = le32(0)
 		i.sb = &sbRef{name: r.name, addend: r.addend}
 		return nil
+	case TLSMem:
+		// off(TLS): the segment-prefixed absolute access, mod=00 with the
+		// SIB escape's disp32 absolute form.  The displacement is the TLS
+		// slot offset, patched by the linker's TLS relocation.
+		i.prefix = r.Seg
+		i.modrm = 0x04 | regField<<3
+		i.sib = 0x25
+		i.disp = le32(r.Disp)
+		i.tls = true
+		return nil
+	case SegAbs:
+		// 0x30(GS): the segment override with the SIB escape's disp32
+		// absolute form, no relocation.
+		setSegAbs(i, regField, r)
+		return nil
 	default:
 		return fmt.Errorf("invalid r/m operand %T", rm)
 	}
 }
 
 func setMem(i *instr, regField int, m Mem) error {
+	if m.Seg != 0 {
+		i.prefix = m.Seg
+	}
 	modrm, sib, disp, xBit, bBit, err := memComponents(regField, m)
 	if err != nil {
 		return err
@@ -692,6 +715,16 @@ func setMem(i *instr, regField int, m Mem) error {
 	return nil
 }
 
+// setSegAbs assembles a segment-absolute operand, 0x30(GS): the segment
+// override with the mod=00 SIB escape's disp32 absolute form and no
+// relocation.
+func setSegAbs(i *instr, regField int, m SegAbs) {
+	i.prefix = m.Seg
+	i.modrm = 0x04 | regField<<3
+	i.sib = 0x25
+	i.disp = le32(m.Disp)
+}
+
 // memComponents computes the ModR/M byte (with the given reg field), the SIB
 // byte (-1 if none), the displacement bytes, and the high index/base bits, for
 // a memory operand.  It is shared by the REX (scalar) and VEX (vector) paths.

asm/encode_test.go

@@ -10,8 +10,8 @@ import (
 
 	"golang.org/x/arch/x86/x86asm"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/ast"
-	"sourcedock.dev/petrbalvin/gasm-devkit/parser"
+	"sourcedock.dev/petrbalvin/gasm-sdk/ast"
+	"sourcedock.dev/petrbalvin/gasm-sdk/parser"
 )
 
 // decode encodes an instruction and decodes it back, returning the decoded

asm/goobj_test.go

@@ -12,7 +12,7 @@ import (
 	"strings"
 	"testing"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/parser"
+	"sourcedock.dev/petrbalvin/gasm-sdk/parser"
 )
 
 // goobjView is a minimal parsed view of a GOOBJ payload, enough to check

asm/guard_test.go

@@ -9,8 +9,8 @@ import (
 	"strings"
 	"testing"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/ast"
-	"sourcedock.dev/petrbalvin/gasm-devkit/parser"
+	"sourcedock.dev/petrbalvin/gasm-sdk/ast"
+	"sourcedock.dev/petrbalvin/gasm-sdk/parser"
 )
 
 // The expected bytes are pinned from `go tool asm` output (Go 1.27, amd64,
@@ -307,12 +307,12 @@ func TestStackGuardBytesLOONG64(t *testing.T) {
 func TestStackGuardGOObjInternalCall(t *testing.T) {
 	for _, tt := range []struct {
 		src      string
-		assemble func(*ast.File) (*Image, error)
+		assemble func(*ast.File, ...AssembleOption) (*Image, error)
 	}{
 		{"g_amd64.s", AssembleFile},
-		{"g_arm64.s", AssembleFileARM64},
-		{"g_riscv64.s", AssembleFileRISCV},
-		{"g_loong64.s", AssembleFileLOONG64},
+		{"g_arm64.s", func(f *ast.File, _ ...AssembleOption) (*Image, error) { return AssembleFileARM64(f) }},
+		{"g_riscv64.s", func(f *ast.File, _ ...AssembleOption) (*Image, error) { return AssembleFileRISCV(f) }},
+		{"g_loong64.s", func(f *ast.File, _ ...AssembleOption) (*Image, error) { return AssembleFileLOONG64(f) }},
 	} {
 		f, errs := parser.Parse(tt.src, "TEXT \u00b7callsmall(SB), $16-0\n\tCALL \u00b7other(SB)\n\tRET\nTEXT \u00b7other(SB), NOSPLIT, $0\n\tRET\n")
 		if len(errs) > 0 {

asm/instrs.go

@@ -192,6 +192,30 @@ func (e *enc) encodeMov(ops []Operand, size int) error {
 		}
 		return e.emit(i)
 
+	case TLSMem:
+		if !dstIsReg {
+			return fmt.Errorf("MOV: two memory operands")
+		}
+		// MOV r, off(TLS): the segment-prefixed absolute load, reg=dst,
+		// rm=src(tlsMem) through the SIB escape; the disp32 is the TLS slot
+		// offset with its R_TLSLE patch site.
+		i := newInstr(size, []byte{movRR(size)})
+		if err := setRM(i, dstReg, src, size); err != nil {
+			return err
+		}
+		return e.emit(i)
+
+	case SegAbs:
+		if !dstIsReg {
+			return fmt.Errorf("MOV: two memory operands")
+		}
+		// MOV r, 0x30(GS): the segment-absolute load.
+		i := newInstr(size, []byte{movRR(size)})
+		if err := setRM(i, dstReg, src, size); err != nil {
+			return err
+		}
+		return e.emit(i)
+
 	case Imm:
 		if dstIsReg {
 			v := int64(src)
@@ -232,11 +256,24 @@ func (e *enc) encodeMov(ops []Operand, size int) error {
 			i.imm = imm
 			return e.emit(i)
 		}
-		// MOV r/m, imm: 0xC6 (8-bit) / 0xC7 /0.
+		// MOV r/m, imm: 0xC6 (8-bit) / 0xC7 /0.  An immediate in the
+		// destination slot is the absolute-address crash-store spelling,
+		// MOVL $0xf1, 0xf1: the parser reads the trailing bare constant
+		// as an immediate, and the store's disp32 carries the address.
 		op := byte(0xC7)
 		if size == 1 {
 			op = 0xC6
 		}
+		if d, ok := dst.(Imm); ok {
+			i := newInstr(size, []byte{op})
+			setSegAbs(i, 0, SegAbs{Disp: int64(d)})
+			immBytes, err := immediate(int64(src), size, false)
+			if err != nil {
+				return err
+			}
+			i.imm = immBytes
+			return e.emit(i)
+		}
 		i := newInstr(size, []byte{op})
 		if err := setRMDigit(i, 0, dst, size); err != nil {
 			return err
@@ -627,7 +664,16 @@ func (e *enc) encodeDoubleShift(base string, ops []Operand, size int) error {
 func (e *enc) encodeImul(ops []Operand, size int) error {
 	switch len(ops) {
 	case 2:
-		// IMUL r, r/m: 0x0F 0xAF.
+		// Two shapes. The leading-immediate spelling IMUL $imm, r multiplies
+		// r in place (dst = rm = r): the shape GOROOT's clock code writes.
+		// Otherwise IMUL r, r/m: 0x0F 0xAF.
+		if imm, ok := ops[0].(Imm); ok {
+			dstReg, isReg := ops[1].(Reg)
+			if !isReg {
+				return fmt.Errorf("IMUL: destination must be a register")
+			}
+			return e.encodeImulImm(imm, dstReg, dstReg, size)
+		}
 		dstReg, ok := ops[1].(Reg)
 		if !ok {
 			return fmt.Errorf("IMUL: destination must be a register")
@@ -647,17 +693,26 @@ func (e *enc) encodeImul(ops []Operand, size int) error {
 		if !ok {
 			return fmt.Errorf("IMUL: immediate operand expected first")
 		}
-		// Plan 9 order: IMUL $imm, src, dst.
+		// Plan 9 order: IMUL $imm, src, dst; the source stays a general
+		// r/m operand (setRM takes registers and memory alike).
+		return e.encodeImulImm(imm, ops[1], dstReg, size)
+	}
+	return fmt.Errorf("IMUL expects 2 or 3 operands, got %d", len(ops))
+}
+
+// encodeImulImm emits the immediate multiply: 0x6B with a sign-extended imm8
+// when the value fits, 0x69 with a 32-bit immediate otherwise.
+func (e *enc) encodeImulImm(imm Imm, rm Operand, dst Reg, size int) error {
 	if fits8(int64(imm)) {
 		i := newInstr(size, []byte{0x6B})
-			if err := setRM(i, dstReg, ops[1], size); err != nil {
+		if err := setRM(i, dst, rm, size); err != nil {
 			return err
 		}
 		i.imm = []byte{byte(int8(imm))}
 		return e.emit(i)
 	}
 	i := newInstr(size, []byte{0x69})
-		if err := setRM(i, dstReg, ops[1], size); err != nil {
+	if err := setRM(i, dst, rm, size); err != nil {
 		return err
 	}
 	immBytes, err := immediate(int64(imm), size, false)
@@ -667,8 +722,6 @@ func (e *enc) encodeImul(ops []Operand, size int) error {
 	i.imm = immBytes
 	return e.emit(i)
 }
-	return fmt.Errorf("IMUL expects 2 or 3 operands, got %d", len(ops))
-}
 
 // --- PUSH / POP -------------------------------------------------------------
 

asm/integration_test.go

@@ -16,7 +16,7 @@ import (
 
 	"golang.org/x/arch/x86/x86asm"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/parser"
+	"sourcedock.dev/petrbalvin/gasm-sdk/parser"
 )
 
 // TestAssembleGoFlacAVX2Kernel assembles the whole production AVX2 kernel;
@@ -25,7 +25,7 @@ import (
 func TestAssembleGoFlacAVX2Kernel(t *testing.T) {
 	path := "../../go-libraries/go-flac/avx2_amd64.s"
 	if _, err := os.Stat(path); err != nil {
-		t.Skip("go-libraries repository not present next to gasm-devkit")
+		t.Skip("go-libraries repository not present next to gasm-sdk")
 	}
 	src, err := os.ReadFile(path)
 	if err != nil {
@@ -86,7 +86,7 @@ func TestAssembleGoFlacAVX2Kernel(t *testing.T) {
 func TestAssembleGoFlacAVX512Kernel(t *testing.T) {
 	path := "../../go-libraries/go-flac/avx512_amd64.s"
 	if _, err := os.Stat(path); err != nil {
-		t.Skip("go-libraries repository not present next to gasm-devkit")
+		t.Skip("go-libraries repository not present next to gasm-sdk")
 	}
 	src, err := os.ReadFile(path)
 	if err != nil {

asm/kernels_differential_test.go

@@ -13,7 +13,7 @@ import (
 	"strings"
 	"testing"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/parser"
+	"sourcedock.dev/petrbalvin/gasm-sdk/parser"
 )
 
 // The differential kernels for the DATA-path and front-end gaps are kept in
@@ -130,6 +130,7 @@ func TestDifferentialKernels(t *testing.T) {
 		{filepath.Join("..", "testdata", "verify", "quadreg_amd64.s"), "", false},
 		{filepath.Join("..", "testdata", "verify", "floatimm_amd64.s"), "", false},
 		{filepath.Join("..", "testdata", "verify", "bookkeep_amd64.s"), "", false},
+		{filepath.Join("..", "testdata", "verify", "forms_amd64.s"), "", false},
 		{filepath.Join("..", "testdata", "verify", "datarel_arm64.s"), "arm64", true},
 		{filepath.Join("..", "testdata", "verify", "divslash_arm64.s"), "arm64", true},
 	} {

asm/l64_goobj_test.go

@@ -12,7 +12,7 @@ import (
 	"strings"
 	"testing"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/parser"
+	"sourcedock.dev/petrbalvin/gasm-sdk/parser"
 )
 
 // TestGOObjectLOONG64Structure checks the emitted loong64 object's blocks:

asm/link.go

@@ -9,7 +9,7 @@ import (
 	"sort"
 	"strconv"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/ast"
+	"sourcedock.dev/petrbalvin/gasm-sdk/ast"
 )
 
 // Image is an assembled file: the function bodies laid out in source order,
@@ -150,6 +150,18 @@ func (img *Image) Bytes() []byte {
 	return append(out, img.Data...)
 }
 
+// AssembleOption adjusts the file-level assembly context.
+type AssembleOption func(*linkInfo)
+
+// WithGOOS selects the target operating system for the forms that depend on
+// it, the TLS access shape above all: linux and freebsd take the
+// one-instruction form, windows and plan9 keep the two-instruction load.
+func WithGOOS(goos string) AssembleOption {
+	return func(l *linkInfo) {
+		l.goos = goos
+	}
+}
+
 // AssembleFile assembles every TEXT function of a parsed file and lays out
 // its static symbols (GLOBL/DATA) in a data section behind the code.  Each
 // reference to a file-local static symbol becomes a RIP-relative load whose
@@ -157,7 +169,7 @@ func (img *Image) Bytes() []byte {
 // GLOBL defines is recorded as an external relocation (Externals) with its
 // displacement left zero, the object-file emitters resolve it at link
 // time, while the raw image (Bytes) cannot represent it.
-func AssembleFile(f *ast.File) (*Image, error) {
+func AssembleFile(f *ast.File, opts ...AssembleOption) (*Image, error) {
 	dataSyms, err := collectData(f)
 	if err != nil {
 		return nil, err
@@ -166,7 +178,17 @@ func AssembleFile(f *ast.File) (*Image, error) {
 	for _, d := range dataSyms {
 		known[d.name] = true
 	}
+	// TEXT symbols are file-level definitions too: a symbol immediate
+	// ($fn(SB)) may name one, exactly as a data reference names a GLOBL.
+	for _, d := range f.Decls {
+		if t, ok := d.(*ast.Text); ok {
+			known[t.Name.Name] = true
+		}
+	}
 	link := &linkInfo{symbols: known, allowExternal: true}
+	for _, o := range opts {
+		o(link)
+	}
 	poolSeen := map[string]bool{}
 
 	img := &Image{Symbols: map[string]int{}, SourcePath: f.Path}

asm/link_test.go

@@ -12,7 +12,7 @@ import (
 	"strings"
 	"testing"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/parser"
+	"sourcedock.dev/petrbalvin/gasm-sdk/parser"
 )
 
 // TestAssembleFileStaticData checks the whole-image layout; code, padding

asm/loong64_assemble.go

@@ -9,7 +9,7 @@ import (
 	"strconv"
 	"strings"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/ast"
+	"sourcedock.dev/petrbalvin/gasm-sdk/ast"
 )
 
 // assembleLOONG64 assembles a LoongArch (loong64) TEXT function body into

asm/loong64_encode_test.go

@@ -8,8 +8,8 @@ import (
 	"encoding/binary"
 	"testing"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/ast"
-	"sourcedock.dev/petrbalvin/gasm-devkit/parser"
+	"sourcedock.dev/petrbalvin/gasm-sdk/ast"
+	"sourcedock.dev/petrbalvin/gasm-sdk/parser"
 )
 
 // firstTextLOONG64 parses assembly source and returns the first TEXT body.

asm/loong64_frame.go

@@ -6,7 +6,7 @@ package asm
 import (
 	"strings"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/ast"
+	"sourcedock.dev/petrbalvin/gasm-sdk/ast"
 )
 
 // Loong64 frame mapping, matching the Go toolchain's loong64 backend.

asm/loong64_more_test.go

@@ -7,7 +7,7 @@ import (
 	"bytes"
 	"testing"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/parser"
+	"sourcedock.dev/petrbalvin/gasm-sdk/parser"
 )
 
 // TestLOONG64_sys exercises the no-operand system instructions and the

asm/loong64_reloc_test.go

@@ -6,7 +6,7 @@ package asm
 import (
 	"testing"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/parser"
+	"sourcedock.dev/petrbalvin/gasm-sdk/parser"
 )
 
 // TestLOONG64RelocOffsetsIncludePrologue pins the function-relative

asm/operand.go

@@ -36,6 +36,29 @@ type FloatImm struct {
 
 func (FloatImm) isOperand() {}
 
+// TLSMem is a thread-local access, the source form off(base)(TLS*1) with the
+// base dropped: the toolchain's one-instruction TLS rewrite assembles it as
+// the segment-prefixed absolute whose disp32 carries an R_TLS_LE patch site
+// (the linker fills the TLS slot offset).
+type TLSMem struct {
+	Disp int64
+	Size int
+	Seg  byte // the segment override: FS (0x64) or GS (0x65) on windows
+}
+
+func (TLSMem) isOperand() {}
+
+// SegAbs is a segment-absolute access, 0x30(GS): the segment override
+// prefixes a disp32 absolute reference with no relocation.  The base
+// register spellings GS and FS produce it.
+type SegAbs struct {
+	Disp int64
+	Size int
+	Seg  byte // 0x64 FS, 0x65 GS
+}
+
+func (SegAbs) isOperand() {}
+
 // Mem is a memory operand of the form disp(base)(index*scale).
 type Mem struct {
 	Base     Reg
@@ -45,6 +68,7 @@ type Mem struct {
 	Size     int // operand width in bytes
 	HasBase  bool
 	HasIndex bool
+	Seg      byte // segment override prefix (0x64 FS, 0x65 GS); 0 = none
 }
 
 func (Mem) isOperand() {}

asm/riscv_assemble.go

@@ -10,7 +10,7 @@ import (
 	"slices"
 	"strings"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/ast"
+	"sourcedock.dev/petrbalvin/gasm-sdk/ast"
 )
 
 // assembleRISCV assembles a RISC-V TEXT function body into machine code.

asm/riscv_encode_test.go

@@ -10,8 +10,8 @@ import (
 	"strings"
 	"testing"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/ast"
-	"sourcedock.dev/petrbalvin/gasm-devkit/parser"
+	"sourcedock.dev/petrbalvin/gasm-sdk/ast"
+	"sourcedock.dev/petrbalvin/gasm-sdk/parser"
 )
 
 // firstTextRISCV parses assembly source and returns the first TEXT function body.

asm/riscv_frame.go

@@ -7,7 +7,7 @@ import (
 	"fmt"
 	"strings"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/ast"
+	"sourcedock.dev/petrbalvin/gasm-sdk/ast"
 )
 
 // RISC-V frame mapping, matching the Go toolchain's riscv64 backend.

asm/riscv_frame_test.go

@@ -6,7 +6,7 @@ package asm
 import (
 	"testing"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/parser"
+	"sourcedock.dev/petrbalvin/gasm-sdk/parser"
 )
 
 // TestRISCVFrameSpadjAndLines checks that a framed function records its

asm/riscv_goobj_test.go

@@ -13,7 +13,7 @@ import (
 	"strings"
 	"testing"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/parser"
+	"sourcedock.dev/petrbalvin/gasm-sdk/parser"
 )
 
 // TestGOObjectRISCVCallReloc checks that CALL sym(SB) emits a single JAL

ast/ast.go

@@ -8,7 +8,7 @@
 // to the arch package; the AST records syntax only.
 package ast
 
-import "sourcedock.dev/petrbalvin/gasm-devkit/token"
+import "sourcedock.dev/petrbalvin/gasm-sdk/token"
 
 // File is the parsed representation of one .s source file.
 type File struct {

ast/ast_test.go

@@ -6,7 +6,7 @@ package ast
 import (
 	"testing"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/token"
+	"sourcedock.dev/petrbalvin/gasm-sdk/token"
 )
 
 func pos(line, col int) token.Position { return token.Position{Line: line, Column: col} }

cmd/gasm/asmhdr.go

@@ -17,7 +17,7 @@ import (
 	"regexp"
 	"strings"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/arch"
+	"sourcedock.dev/petrbalvin/gasm-sdk/arch"
 )
 
 // go_asm.h is the header the Go compiler writes for every package that

cmd/gasm/asmhdr_test.go

@@ -5,6 +5,7 @@ package main
 
 import (
 	"os"
+	"os/exec"
 	"path/filepath"
 	"strings"
 	"testing"
@@ -392,14 +393,67 @@ func TestRunCorpusAuditGOOS(t *testing.T) {
 	}
 }
 
+// TestRunCorpusAuditBuildConstraint covers the //go:build classification end
+// to end: a generic-named file whose constraint admits one target is
+// attempted there alone (cpu_x86.s on amd64), and a file whose constraint
+// admits none of the four targets is never attempted (the msan and
+// goexperiment trees).
+func TestRunCorpusAuditBuildConstraint(t *testing.T) {
+	dir := t.TempDir()
+	write := func(name, src string) {
+		t.Helper()
+		if err := os.WriteFile(filepath.Join(dir, name), []byte(src), 0o644); err != nil {
+			t.Fatal(err)
+		}
+	}
+	write("x86.s", "//go:build 386 || amd64\n\nTEXT \xc2\xb7f(SB), NOSPLIT, $0\n\tRET\n")
+	write("racey.s", "//go:build race\n\nTEXT \xc2\xb7r(SB), NOSPLIT, $0\n\tRET\n")
+	write("plain.s", "TEXT \xc2\xb7p(SB), NOSPLIT, $0\n\tRET\n")
+
+	stats, err := runCorpusAudit(dir, nil)
+	if err != nil {
+		t.Fatalf("runCorpusAudit: %v", err)
+	}
+	tally := func(name string) *corpusTally {
+		for i, tg := range stats.targets {
+			if tg.name == name {
+				return stats.tallies[i]
+			}
+		}
+		t.Fatalf("no tally for %s", name)
+		return nil
+	}
+	if stats.narrowed != 1 || stats.excluded != 1 || stats.generic != 1 {
+		t.Errorf("buckets = narrowed %d, excluded %d, generic %d; want 1, 1, 1", stats.narrowed, stats.excluded, stats.generic)
+	}
+	if a := tally("amd64"); a.attempted != 2 || a.assembled != 2 {
+		t.Errorf("amd64 = %d/%d, want 2/2 (x86.s and plain.s)", a.assembled, a.attempted)
+	}
+	for _, name := range []string{"arm64", "riscv64", "loong64"} {
+		if a := tally(name); a.attempted != 1 || a.assembled != 1 {
+			t.Errorf("%s = %d/%d, want 1/1 (plain.s only)", name, a.assembled, a.attempted)
+		}
+	}
+	if stats.full != 2 {
+		t.Errorf("full = %d, want 2 (x86.s over its one target, plain.s over all four)", stats.full)
+	}
+}
+
 // TestGenerateGoAsmHeaderRuntime pins the generator against the real thing:
-// the runtime package, whose header the toolchain's own -asmhdr output was
-// sampled from.  Skipped in short mode: it type-checks the whole package.
+// the runtime package of the ambient toolchain, whose header the toolchain's
+// own -asmhdr output was sampled from.  Skipped in short mode: it type-checks
+// the whole package.  The GOROOT comes from the go command itself, so the
+// test follows whatever toolchain the host provides.
 func TestGenerateGoAsmHeaderRuntime(t *testing.T) {
 	if testing.Short() {
 		t.Skip("type-checks the whole runtime package")
 	}
-	dir, err := generateGoAsmHeader("/usr/local/go/src/runtime", "", "amd64", t.TempDir())
+	out, err := exec.Command("go", "env", "GOROOT").Output()
+	if err != nil {
+		t.Skipf("no Go toolchain: %v", err)
+	}
+	runtimeDir := filepath.Join(strings.TrimSpace(string(out)), "src", "runtime")
+	dir, err := generateGoAsmHeader(runtimeDir, "", "amd64", t.TempDir())
 	if err != nil {
 		t.Fatalf("generateGoAsmHeader(runtime): %v", err)
 	}

cmd/gasm/audit.go

@@ -5,6 +5,7 @@ package main
 
 import (
 	"fmt"
+	"go/build/constraint"
 	"maps"
 	"os"
 	"os/exec"
@@ -15,9 +16,9 @@ import (
 	"strconv"
 	"strings"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/arch"
-	"sourcedock.dev/petrbalvin/gasm-devkit/asm"
-	"sourcedock.dev/petrbalvin/gasm-devkit/parser"
+	"sourcedock.dev/petrbalvin/gasm-sdk/arch"
+	"sourcedock.dev/petrbalvin/gasm-sdk/asm"
+	"sourcedock.dev/petrbalvin/gasm-sdk/parser"
 )
 
 // cmdAuditInstructions cross-checks a gasm encoder against the Go toolchain's
@@ -38,7 +39,7 @@ import (
 // construction and are excluded from the diff; the other architectures list
 // their conditional branches outright.
 func cmdAuditInstructions(args []string) error {
-	fs := newCommand("audit-instructions", "gasm audit-instructions [--corpus [dir]] [-I dir] [amd64|arm64|riscv64|loong64]", `
+	fs := newCommand("audit-instructions", "gasm audit-instructions [--corpus [dir]] [--list] [-I dir] [amd64|arm64|riscv64|loong64]", `
 Compare the gasm encoder for the given architecture (default amd64) against
 go tool asm and print the diff: superset encodings (gasm-only, shippable via
 gasm asm --format goobj) and known-but-unencodable names (the backlog).  The
@@ -55,16 +56,19 @@ toolchain probing.  A file whose name carries a recognisable _arch suffix is
 attempted for that architecture; a file without one is attempted for all
 four, exactly as a GOARCH build would compile it.  The report gives the
 per-architecture pass rates and the most common failure reasons, which drive
-the encodability backlog by frequency rather than by table order.
+the encodability backlog by frequency rather than by table order.  With
+-list the report also prints every failing file with its reason, per
+architecture.
 `)
 	corpus := fs.Bool("corpus", false, "assemble a corpus of .s files and report pass rates and failure reasons")
+	list := fs.Bool("list", false, "with --corpus, list every failing file with its reason, per architecture")
 	var dirs includeDirs
 	fs.Var(&dirs, "I", "directory to search for #include files (may be repeated)")
 	if err := fs.Parse(args); err != nil {
 		return err
 	}
 	if *corpus {
-		return cmdAuditCorpus(fs.Args(), dirs)
+		return cmdAuditCorpus(fs.Args(), dirs, *list)
 	}
 	archName := "amd64"
 	switch n := len(fs.Args()); {
@@ -403,20 +407,30 @@ type corpusTally struct {
 	assembled int
 	reasons   map[string]int    // failure reason → count
 	example   map[string]string // failure reason → one representative file
+	fails     []corpusFailure   // every failure, in file order, for --list
 }
 
-func (t *corpusTally) fail(path, reason string) {
+// corpusFailure is one failed attempt, recorded for the --list report.
+type corpusFailure struct {
+	path   string
+	reason string
+	detail string
+}
+
+func (t *corpusTally) fail(path string, err error) {
+	reason := corpusReason(err)
 	t.reasons[reason]++
 	if t.example[reason] == "" {
 		t.example[reason] = path
 	}
+	t.fails = append(t.fails, corpusFailure{path: path, reason: reason, detail: firstLine(err.Error())})
 }
 
 // cmdAuditCorpus implements audit-instructions --corpus.  The include
 // directories carry #include resolution over a corpus whose files refer to
 // headers such as GOROOT/pkg/include, the same -I a toolchain comparison
 // needs.
-func cmdAuditCorpus(args []string, dirs includeDirs) error {
+func cmdAuditCorpus(args []string, dirs includeDirs, list bool) error {
 	if len(args) > 1 {
 		return &usageError{fmt.Errorf("audit-instructions --corpus takes at most one directory argument")}
 	}
@@ -454,7 +468,7 @@ func cmdAuditCorpus(args []string, dirs includeDirs) error {
 	if err != nil {
 		return err
 	}
-	printCorpusStats(stats)
+	printCorpusStats(stats, list)
 	return nil
 }
 
@@ -463,8 +477,10 @@ type corpusStats struct {
 	root      string
 	files     int
 	generic   int // files attempted for all four architectures
+	narrowed  int // files whose //go:build admits a proper subset of the four
+	excluded  int // files whose //go:build admits none of the four: never compiled
 	otherPort int // files named for another Go port: never attempted
-	full      int // files that assembled for every target architecture
+	full      int // files that assembled for every applicable target architecture
 	targets   []corpusTarget
 	tallies   []*corpusTally
 }
@@ -560,6 +576,57 @@ func otherGOOSFile(path string) bool {
 	return false
 }
 
+// buildConstraint returns the file's leading //go:build expression, or nil
+// when the file carries none.  The constraint governs the same header block
+// go/build reads: blank lines and comments may precede it, and the first
+// line that is neither ends the block.  A constraint that does not parse
+// narrows nothing, so the file stays in the attempted set: the audit must
+// never exclude a file the toolchain would compile.
+func buildConstraint(src string) constraint.Expr {
+	for line := range strings.SplitSeq(src, "\n") {
+		t := strings.TrimSpace(line)
+		switch {
+		case t == "":
+			continue
+		case strings.HasPrefix(t, "//"):
+			if constraint.IsGoBuild(t) {
+				e, err := constraint.Parse(t)
+				if err != nil {
+					return nil
+				}
+				return e
+			}
+			continue
+		default:
+			return nil
+		}
+	}
+	return nil
+}
+
+// unixOS is go/build's unixOS set: the GOOSes the unix build tag admits.
+var unixOS = map[string]bool{
+	"aix": true, "android": true, "darwin": true, "dragonfly": true,
+	"freebsd": true, "hurd": true, "illumos": true, "ios": true,
+	"linux": true, "netbsd": true, "openbsd": true, "solaris": true,
+}
+
+// constraintTags answers the build tags a plain `go build` sets for a
+// target: the GOOS and GOARCH, gc, and unix on the unix-like GOOSes.  No
+// experiment, sanitiser or cgo tag is ever true: the audit models the
+// default build, and no GOROOT assembly file's constraint hinges on cgo.
+func constraintTags(goarch, goos string) func(string) bool {
+	return func(tag string) bool {
+		switch tag {
+		case goarch, goos, "gc":
+			return true
+		case "unix":
+			return unixOS[goos]
+		}
+		return false
+	}
+}
+
 func runCorpusAudit(root string, dirs includeDirs) (*corpusStats, error) {
 	files, err := asmFiles(root)
 	if err != nil {
@@ -577,8 +644,8 @@ func runCorpusAudit(root string, dirs includeDirs) (*corpusStats, error) {
 		tallies[i] = &corpusTally{reasons: map[string]int{}, example: map[string]string{}}
 	}
 	// full is the north-star number: a file counts when every architecture
-	// its name allows assembles it.
-	full, generic, otherPort := 0, 0, 0
+	// its build admits assembles it.
+	full, generic, otherPort, narrowedCount, excluded := 0, 0, 0, 0, 0
 
 	// Header generation is created on first use, so a corpus with no
 	// go_asm.h includes never pays for a temp directory.
@@ -601,28 +668,78 @@ func runCorpusAudit(root string, dirs includeDirs) (*corpusStats, error) {
 		// invisible to a file-name rule).
 		goos := goosFromFilename(path)
 
+		// The GOOS the header generation type-checks under follows the
+		// file's name when the name carries one; the ambient GOOS is the
+		// honest guess otherwise.
+		namedArch := arch.FromFilename(path)
 		var wanted []int // indexes into targets
-		if a := arch.FromFilename(path); a != arch.Unknown {
+		other := false
+		switch {
+		case namedArch != arch.Unknown:
 			for i, tg := range targets {
-				if tg.a == a {
+				if tg.a == namedArch {
 					wanted = append(wanted, i)
 				}
 			}
-		} else if otherPortFile(path) {
+		case otherPortFile(path):
 			// A file named for a Go port gasm does not support (arm,
 			// 386, s390x, ...) or for another GOOS is compiled by no
 			// supported-arch build, so it is neither generic nor a
 			// per-arch attempt: counting it as generic would make the
 			// headline unreachably low for reasons no supported target
 			// can fix.
+			other = true
 			otherPort++
-		} else {
-			generic++
+		default:
 			for i := range targets {
 				wanted = append(wanted, i)
 			}
 		}
 
+		// A //go:build constraint narrows the set of targets the file is
+		// assembled for, the way the go command compiles the file only for
+		// the targets the expression admits: cpu_x86.s belongs to the x86
+		// build alone, and a file whose constraint admits none of the four
+		// targets (the goexperiment.runtimesecret and msan trees) is
+		// compiled by no supported build.  The tags mirror what a plain
+		// `go build` sets: the GOOS and GOARCH, gc, and unix on the
+		// unix-like GOOSes; no experiment, sanitiser or cgo tag is ever
+		// true.  The GOOS is the file's own when the name carries one,
+		// else the ambient one.
+		goosForEval := goos
+		if goosForEval == "" {
+			goosForEval = runtime.GOOS
+		}
+		narrowed := false
+		if len(wanted) > 0 {
+			if ce := buildConstraint(src); ce != nil {
+				kept := make([]int, 0, len(wanted))
+				for _, i := range wanted {
+					tg := targets[i]
+					if ce.Eval(constraintTags(goarchName(tg.a), goosForEval)) {
+						kept = append(kept, i)
+					}
+				}
+				if len(kept) < len(wanted) {
+					narrowed = true
+				}
+				wanted = kept
+			}
+		}
+
+		switch {
+		case other:
+			// already tallied above
+		case len(wanted) == 0:
+			excluded++
+		case namedArch != arch.Unknown:
+			// a per-arch attempt over the constraint's subset
+		case narrowed:
+			narrowedCount++
+		default:
+			generic++
+		}
+
 		// A file that includes go_asm.h parses against a per-target header:
 		// the defines differ per architecture (internal/cpu's layout, for
 		// one) and per GOOS (sys_darwin_arm64.s's trampoline constants,
@@ -645,21 +762,22 @@ func runCorpusAudit(root string, dirs includeDirs) (*corpusStats, error) {
 				hdrDir, err := hdr.dirFor(pkgDir, goos, goarchName(tg.a))
 				if err != nil {
 					ok = false
-					t.fail(path, corpusReason(err))
+					t.fail(path, err)
 					continue
 				}
 				f, errs := parser.ParseWithOptions(path, src, parser.Options{
 					Expand:      true,
 					IncludeDirs: append(slices.Clone(dirs), hdrDir),
+					Predefines:  platformPredefinesFor(goarchName(tg.a), goos),
 				})
 				if len(errs) > 0 {
 					ok = false
-					t.fail(path, corpusReason(errs[0]))
+					t.fail(path, errs[0])
 					continue
 				}
-				if _, err := assembleFile(tg.a, f); err != nil {
+				if _, err := assembleFile(tg.a, f, goos); err != nil {
 					ok = false
-					t.fail(path, corpusReason(err))
+					t.fail(path, err)
 					continue
 				}
 				t.assembled++
@@ -670,21 +788,28 @@ func runCorpusAudit(root string, dirs includeDirs) (*corpusStats, error) {
 			continue
 		}
 
-		f, errs := parser.ParseWithOptions(path, src, parser.Options{Expand: true, IncludeDirs: dirs})
-
 		ok := true
 		for _, i := range wanted {
 			tg, t := targets[i], tallies[i]
 			t.attempted++
+			// The parse carries the target's platform predefines, so it
+			// cannot be shared across targets the way a header-free file's
+			// could: a #ifdef GOARCH_arm block must be live on arm64 and
+			// dead everywhere else.
+			f, errs := parser.ParseWithOptions(path, src, parser.Options{
+				Expand:      true,
+				IncludeDirs: dirs,
+				Predefines:  platformPredefinesFor(goarchName(tg.a), goos),
+			})
 			var err error
 			if len(errs) > 0 {
 				err = errs[0] // a parse failure is a failure for every target
 			} else {
-				_, err = assembleFile(tg.a, f)
+				_, err = assembleFile(tg.a, f, goos)
 			}
 			if err != nil {
 				ok = false
-				t.fail(path, corpusReason(err))
+				t.fail(path, err)
 				continue
 			}
 			t.assembled++
@@ -698,6 +823,8 @@ func runCorpusAudit(root string, dirs includeDirs) (*corpusStats, error) {
 		root:      root,
 		files:     len(files),
 		generic:   generic,
+		narrowed:  narrowedCount,
+		excluded:  excluded,
 		otherPort: otherPort,
 		full:      full,
 		targets:   targets,
@@ -706,14 +833,16 @@ func runCorpusAudit(root string, dirs includeDirs) (*corpusStats, error) {
 }
 
 // printCorpusStats renders the corpus audit report.
-func printCorpusStats(s *corpusStats) {
-	fmt.Printf("corpus %s: %d files (%d generic, attempted for all architectures; %d named for other Go ports, never attempted)\n", s.root, s.files, s.generic, s.otherPort)
+func printCorpusStats(s *corpusStats, list bool) {
+	fmt.Printf("corpus %s: %d files (%d generic, attempted for all architectures; %d narrowed by //go:build; %d excluded by //go:build; %d named for other Go ports, never attempted)\n",
+		s.root, s.files, s.generic, s.narrowed, s.excluded, s.otherPort)
 	// The rate is over the files a supported build would attempt: the
-	// other ports' files sit in the count for completeness but can never
-	// assemble, so counting them in the denominator would report the gap
-	// of architectures gasm deliberately does not target.
-	attemptable := max(s.files-s.otherPort, 1)
-	fmt.Printf("  assemble for every target architecture: %d of %d attemptable (%.1f%%)\n", s.full, attemptable, 100*float64(s.full)/float64(attemptable))
+	// other ports' files and the ones no supported target compiles sit in
+	// the count for completeness but can never assemble, so counting them
+	// in the denominator would report the gap of platforms gasm
+	// deliberately does not target.
+	attemptable := max(s.files-s.otherPort-s.excluded, 1)
+	fmt.Printf("  assemble for every applicable target: %d of %d attemptable (%.1f%%)\n", s.full, attemptable, 100*float64(s.full)/float64(attemptable))
 	for i, tg := range s.targets {
 		t := s.tallies[i]
 		fmt.Printf("  %s: %d/%d attempted\n", tg.name, t.assembled, t.attempted)
@@ -721,6 +850,13 @@ func printCorpusStats(s *corpusStats) {
 			fmt.Printf("    %4d  %s\n", t.reasons[r], r)
 			fmt.Printf("         e.g. %s\n", t.example[r])
 		}
+		if !list {
+			continue
+		}
+		for _, f := range t.fails {
+			fmt.Printf("    FAIL %s\n", f.path)
+			fmt.Printf("         %s: %s\n", f.reason, f.detail)
+		}
 	}
 }
 

cmd/gasm/audit_test.go

@@ -4,9 +4,10 @@
 package main
 
 import (
+	"runtime"
 	"testing"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/arch"
+	"sourcedock.dev/petrbalvin/gasm-sdk/arch"
 )
 
 func TestDerivedFamily(t *testing.T) {
@@ -64,3 +65,62 @@ func TestGasmEncodable(t *testing.T) {
 		}
 	}
 }
+
+// TestBuildConstraint pins the //go:build reader: the constraint governs the
+// leading comment block, the first non-comment line ends it (a tag below a
+// #include governs nothing, exactly as go/build drops it), and a file
+// without one admits every target.
+func TestBuildConstraint(t *testing.T) {
+	admits := func(src, goarch, goos string) bool {
+		t.Helper()
+		e := buildConstraint(src)
+		if e == nil {
+			return true
+		}
+		return e.Eval(constraintTags(goarch, goos))
+	}
+	const ret = "TEXT \xc2\xb7f(SB), NOSPLIT, $0\n\tRET\n"
+	cases := []struct {
+		name         string
+		src          string
+		amd64, arm64 bool
+	}{
+		{"no constraint", ret, true, true},
+		{"x86 only", "//go:build 386 || amd64\n\n" + ret, true, false},
+		{"arm64 and linux", "//go:build arm64 && linux\n\n" + ret, false, true},
+		{"msan never", "//go:build msan\n\n" + ret, false, false},
+		{"experiment never", "//go:build goexperiment.runtimesecret\n\n" + ret, false, false},
+		{"below an include governs nothing", "#include \"textflag.h\"\n//go:build amd64\n" + ret, true, true},
+		{"unparsable narrows nothing", "//go:build (amd64\n" + ret, true, true},
+	}
+	for _, c := range cases {
+		t.Run(c.name, func(t *testing.T) {
+			if got := admits(c.src, "amd64", runtime.GOOS); got != c.amd64 {
+				t.Errorf("amd64 admission = %v, want %v", got, c.amd64)
+			}
+			if got := admits(c.src, "arm64", runtime.GOOS); got != c.arm64 {
+				t.Errorf("arm64 admission = %v, want %v", got, c.arm64)
+			}
+		})
+	}
+}
+
+// TestConstraintTags pins the tag set a plain `go build` sets: the GOOS and
+// GOARCH, gc, unix on the unix-like GOOSes; nothing else is ever true.
+func TestConstraintTags(t *testing.T) {
+	ok := constraintTags("amd64", "linux")
+	for _, tag := range []string{"amd64", "linux", "gc", "unix"} {
+		if !ok(tag) {
+			t.Errorf("tag %q = false, want true", tag)
+		}
+	}
+	for _, tag := range []string{"arm64", "freebsd", "darwin", "cgo", "race", "msan", "goexperiment.runtimesecret"} {
+		if ok(tag) {
+			t.Errorf("tag %q = true, want false", tag)
+		}
+	}
+	fb := constraintTags("arm64", "freebsd")
+	if !fb("unix") {
+		t.Error("unix on freebsd = false, want true")
+	}
+}

cmd/gasm/debug_other.go

@@ -1,7 +1,7 @@
 // Copyright (c) 2026 Petr Balvín <opensource@petrbalvin.org> (https://petrbalvin.org)
 // SPDX-License-Identifier: BSD-3-Clause
 
-//go:build !linux
+//go:build !(linux || (freebsd && (amd64 || arm64 || riscv64)))
 
 package main
 
@@ -11,6 +11,6 @@ import (
 )
 
 func cmdDebug(args []string) int {
-	fmt.Fprintln(os.Stderr, "gasm debug: the interactive debugger requires Linux (ptrace)")
+	fmt.Fprintln(os.Stderr, "gasm debug: the interactive debugger requires Linux or FreeBSD (ptrace)")
 	return 1
 }

cmd/gasm/debug_ptrace.go

@@ -1,7 +1,7 @@
 // Copyright (c) 2026 Petr Balvín <opensource@petrbalvin.org> (https://petrbalvin.org)
 // SPDX-License-Identifier: BSD-3-Clause
 
-//go:build linux
+//go:build linux || (freebsd && (amd64 || arm64 || riscv64))
 
 package main
 
@@ -13,8 +13,8 @@ import (
 	"strings"
 	"time"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/debug"
-	"sourcedock.dev/petrbalvin/gasm-devkit/verify"
+	"sourcedock.dev/petrbalvin/gasm-sdk/debug"
+	"sourcedock.dev/petrbalvin/gasm-sdk/verify"
 )
 
 func cmdDebug(args []string) int {

cmd/gasm/dis.go

@@ -9,9 +9,9 @@ import (
 	"sort"
 	"strings"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/arch"
-	"sourcedock.dev/petrbalvin/gasm-devkit/disasm"
-	"sourcedock.dev/petrbalvin/gasm-devkit/parser"
+	"sourcedock.dev/petrbalvin/gasm-sdk/arch"
+	"sourcedock.dev/petrbalvin/gasm-sdk/disasm"
+	"sourcedock.dev/petrbalvin/gasm-sdk/parser"
 )
 
 // cmdDis disassembles machine code: either a raw binary (standard input with
@@ -84,7 +84,7 @@ func disSource(path string, target arch.Arch) int {
 	if len(errs) > 0 {
 		return 1
 	}
-	img, err := assembleFile(target, f)
+	img, err := assembleFile(target, f, "")
 	if err != nil {
 		fmt.Fprintf(os.Stderr, "gasm dis: %v\n", err)
 		return 1

cmd/gasm/main.go

@@ -27,15 +27,15 @@ import (
 	"sync"
 	"syscall"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/arch"
-	"sourcedock.dev/petrbalvin/gasm-devkit/asm"
-	"sourcedock.dev/petrbalvin/gasm-devkit/ast"
-	"sourcedock.dev/petrbalvin/gasm-devkit/format"
-	"sourcedock.dev/petrbalvin/gasm-devkit/lexer"
-	"sourcedock.dev/petrbalvin/gasm-devkit/lint"
-	"sourcedock.dev/petrbalvin/gasm-devkit/lsp"
-	"sourcedock.dev/petrbalvin/gasm-devkit/parser"
-	"sourcedock.dev/petrbalvin/gasm-devkit/verify"
+	"sourcedock.dev/petrbalvin/gasm-sdk/arch"
+	"sourcedock.dev/petrbalvin/gasm-sdk/asm"
+	"sourcedock.dev/petrbalvin/gasm-sdk/ast"
+	"sourcedock.dev/petrbalvin/gasm-sdk/format"
+	"sourcedock.dev/petrbalvin/gasm-sdk/lexer"
+	"sourcedock.dev/petrbalvin/gasm-sdk/lint"
+	"sourcedock.dev/petrbalvin/gasm-sdk/lsp"
+	"sourcedock.dev/petrbalvin/gasm-sdk/parser"
+	"sourcedock.dev/petrbalvin/gasm-sdk/verify"
 )
 
 // version reports the release the toolchain recorded for this build: the
@@ -574,7 +574,7 @@ naming the package.
 		defer cleanup()
 		dirs = append(dirs, hdrDir)
 	}
-	f, errs := parser.ParseWithOptions(path, src, parser.Options{Expand: true, IncludeDirs: dirs})
+	f, errs := parser.ParseWithOptions(path, src, parser.Options{Expand: true, IncludeDirs: dirs, Predefines: platformPredefinesFor(string(targetArch), goos)})
 	for _, e := range errs {
 		fmt.Fprintf(os.Stderr, "%s: %v\n", path, e)
 	}
@@ -582,7 +582,7 @@ naming the package.
 		return 1
 	}
 
-	img, err := assembleFile(targetArch, f)
+	img, err := assembleFile(targetArch, f, goos)
 	if err != nil {
 		fmt.Fprintf(os.Stderr, "%s: %v\n", path, err)
 		return 1
@@ -789,11 +789,34 @@ e.g. --map wideCopyAVX2=wideCopyAVX512 pairs the two regardless of suffix.
 	return 1
 }
 
+// platformPredefines mirrors the go command's assembler invocation, which
+// defines GOOS_<goos> and GOARCH_<arch> as -D macros: GOROOT headers
+// (go_tls.h, asm_riscv64.h) select their platform blocks with #ifdef on
+// exactly those names, so an assembler without them cannot see the platform
+// definitions at all.
+func platformPredefines(goarch, goos string) map[string]string {
+	return map[string]string{
+		"GOARCH_" + goarch: "1",
+		"GOOS_" + goos:     "1",
+	}
+}
+
+// platformPredefinesFor resolves the ambient GOOS the way a build would: a
+// file whose name carries one (sys_darwin_arm64.s) is compiled for that GOOS
+// and nothing else.
+func platformPredefinesFor(goarch string, fileGoos string) map[string]string {
+	goos := fileGoos
+	if goos == "" {
+		goos = runtime.GOOS
+	}
+	return platformPredefines(goarch, goos)
+}
+
 // assembleFile assembles a parsed file for the given architecture and returns the image.
-func assembleFile(targetArch arch.Arch, f *ast.File) (*asm.Image, error) {
+func assembleFile(targetArch arch.Arch, f *ast.File, goos string) (*asm.Image, error) {
 	switch targetArch {
 	case arch.AMD64:
-		return asm.AssembleFile(f)
+		return asm.AssembleFile(f, asm.WithGOOS(goos))
 	case arch.RISCV:
 		return asm.AssembleFileRISCV(f)
 	case arch.ARM64:
@@ -813,18 +836,18 @@ func assemblePath(path string, forced arch.Arch, dirs includeDirs) (*asm.Image,
 	if err != nil {
 		return nil, err
 	}
-	f, errs := parser.ParseWithOptions(path, src, parser.Options{Expand: true, IncludeDirs: dirs})
+	target := forced
+	if target == arch.Unknown {
+		target = arch.FromFilename(path)
+	}
+	f, errs := parser.ParseWithOptions(path, src, parser.Options{Expand: true, IncludeDirs: dirs, Predefines: platformPredefinesFor(string(target), "")})
 	for _, e := range errs {
 		fmt.Fprintf(os.Stderr, "%s: %v\n", path, e)
 	}
 	if len(errs) > 0 {
 		return nil, fmt.Errorf("parse errors")
 	}
-	target := forced
-	if target == arch.Unknown {
-		target = arch.FromFilename(path)
-	}
-	return assembleFile(target, f)
+	return assembleFile(target, f, "")
 }
 
 // printByteDiff shows the first few byte differences between two code blocks.
@@ -920,7 +943,7 @@ func cmdVerifyNonJIT(path string, targetArch arch.Arch, groundTruth, profile boo
 	if len(errs) > 0 {
 		return 1
 	}
-	img, err := assembleFile(targetArch, f)
+	img, err := assembleFile(targetArch, f, "")
 	if err != nil {
 		fmt.Fprintf(os.Stderr, "gasm verify: %v\n", err)
 		return 1

cmd/gasm/main_test.go

@@ -14,8 +14,8 @@ import (
 	"syscall"
 	"testing"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/arch"
-	"sourcedock.dev/petrbalvin/gasm-devkit/asm"
+	"sourcedock.dev/petrbalvin/gasm-sdk/arch"
+	"sourcedock.dev/petrbalvin/gasm-sdk/asm"
 )
 
 const clean = "#include \"textflag.h\"\n" +

cmd/gasm/scaffold.go

@@ -11,8 +11,8 @@ import (
 	"os"
 	"strings"
 
-	gasmast "sourcedock.dev/petrbalvin/gasm-devkit/ast"
-	gasmparser "sourcedock.dev/petrbalvin/gasm-devkit/parser"
+	gasmast "sourcedock.dev/petrbalvin/gasm-sdk/ast"
+	gasmparser "sourcedock.dev/petrbalvin/gasm-sdk/parser"
 )
 
 // cmdScaffold generates a differential test skeleton for every kernel in a

debug/breakpoint.go

@@ -1,7 +1,7 @@
 // Copyright (c) 2026 Petr Balvín <opensource@petrbalvin.org> (https://petrbalvin.org)
 // SPDX-License-Identifier: BSD-3-Clause
 
-//go:build linux
+//go:build linux || (freebsd && (amd64 || arm64 || riscv64))
 
 package debug
 

debug/disasm_freebsd_amd64.go

@@ -0,0 +1,56 @@
+// Copyright (c) 2026 Petr Balvín <opensource@petrbalvin.org> (https://petrbalvin.org)
+// SPDX-License-Identifier: BSD-3-Clause
+
+//go:build freebsd && amd64
+
+package debug
+
+import (
+	"fmt"
+	"strings"
+
+	"sourcedock.dev/petrbalvin/gasm-sdk/arch"
+	"sourcedock.dev/petrbalvin/gasm-sdk/disasm"
+)
+
+// Disassemble decodes the instruction at the given address in the debuggee's
+// memory and returns its text representation and length in bytes.
+func (s *Session) Disassemble(addr uint64) (string, int, error) {
+	mem, err := s.ReadMemory(addr, 15)
+	if err != nil {
+		return "", 0, err
+	}
+	ins, err := disasm.Decode(arch.AMD64, mem, addr)
+	if err != nil {
+		return "", 0, err
+	}
+	return ins.Text, ins.Len, nil
+}
+
+// DisassembleN decodes up to n instructions starting at addr and returns
+// them as a formatted string with addresses and byte offsets.
+func (s *Session) DisassembleN(addr uint64, n int) string {
+	var result strings.Builder
+	pc := addr
+	for range n {
+		text, length, err := s.Disassemble(pc)
+		if err != nil {
+			result.WriteString(fmt.Sprintf("  %#08x: <error: %v>\n", pc, err))
+			break
+		}
+		result.WriteString(fmt.Sprintf("  %#08x: %s\n", pc, text))
+		if length == 0 {
+			length = 1
+		}
+		pc += uint64(length)
+	}
+	return result.String()
+}
+
+// isCallInsn reports whether disassembled text (x86asm.IntelSyntax) is a
+// call.  The first token must match exactly: a prefix test would also catch
+// unrelated mnemonics.
+func isCallInsn(text string) bool {
+	m, _, _ := strings.Cut(text, " ")
+	return strings.ToLower(m) == "call"
+}

debug/disasm_freebsd_arm64.go

@@ -0,0 +1,60 @@
+// Copyright (c) 2026 Petr Balvín <opensource@petrbalvin.org> (https://petrbalvin.org)
+// SPDX-License-Identifier: BSD-3-Clause
+
+//go:build freebsd && arm64
+
+package debug
+
+import (
+	"fmt"
+	"strings"
+
+	"sourcedock.dev/petrbalvin/gasm-sdk/arch"
+	"sourcedock.dev/petrbalvin/gasm-sdk/disasm"
+)
+
+// Disassemble decodes the instruction at the given address in the debuggee's
+// memory and returns its text representation and length in bytes.
+func (s *Session) Disassemble(addr uint64) (string, int, error) {
+	mem, err := s.ReadMemory(addr, 4)
+	if err != nil {
+		return "", 0, err
+	}
+	ins, err := disasm.Decode(arch.ARM64, mem, addr)
+	if err != nil {
+		return "", 0, err
+	}
+	return ins.Text, ins.Len, nil
+}
+
+// DisassembleN decodes up to n instructions starting at addr and returns
+// them as a formatted string with addresses and byte offsets.
+func (s *Session) DisassembleN(addr uint64, n int) string {
+	var result strings.Builder
+	pc := addr
+	for range n {
+		text, length, err := s.Disassemble(pc)
+		if err != nil {
+			result.WriteString(fmt.Sprintf("  %#08x: <error: %v>\n", pc, err))
+			break
+		}
+		result.WriteString(fmt.Sprintf("  %#08x: %s\n", pc, text))
+		if length == 0 {
+			length = 1
+		}
+		pc += uint64(length)
+	}
+	return result.String()
+}
+
+// isCallInsn reports whether disassembled text (arm64asm.GoSyntax) is a
+// call.  GoSyntax renders bl as CALL; the native mnemonic is accepted too.
+// The first token must match exactly so branches never match.
+func isCallInsn(text string) bool {
+	m, _, _ := strings.Cut(text, " ")
+	switch strings.ToLower(m) {
+	case "call", "bl":
+		return true
+	}
+	return false
+}

debug/disasm_freebsd_riscv64.go

@@ -0,0 +1,62 @@
+// Copyright (c) 2026 Petr Balvín <opensource@petrbalvin.org> (https://petrbalvin.org)
+// SPDX-License-Identifier: BSD-3-Clause
+
+//go:build freebsd && riscv64
+
+package debug
+
+import (
+	"fmt"
+	"strings"
+
+	"sourcedock.dev/petrbalvin/gasm-sdk/arch"
+	"sourcedock.dev/petrbalvin/gasm-sdk/disasm"
+)
+
+// Disassemble decodes the instruction at the given address in the debuggee's
+// memory and returns its text representation and length in bytes.
+func (s *Session) Disassemble(addr uint64) (string, int, error) {
+	mem, err := s.ReadMemory(addr, 4)
+	if err != nil {
+		return "", 0, err
+	}
+	ins, err := disasm.Decode(arch.RISCV, mem, addr)
+	if err != nil {
+		return "", 0, err
+	}
+	return ins.Text, ins.Len, nil
+}
+
+// DisassembleN decodes up to n instructions starting at addr and returns
+// them as a formatted string with addresses and byte offsets.
+func (s *Session) DisassembleN(addr uint64, n int) string {
+	var result strings.Builder
+	pc := addr
+	for range n {
+		text, length, err := s.Disassemble(pc)
+		if err != nil {
+			result.WriteString(fmt.Sprintf("  %#08x: <error: %v>\n", pc, err))
+			break
+		}
+		result.WriteString(fmt.Sprintf("  %#08x: %s\n", pc, text))
+		if length == 0 {
+			length = 1
+		}
+		pc += uint64(length)
+	}
+	return result.String()
+}
+
+// isCallInsn reports whether disassembled text (riscv64asm.GoSyntax) is a
+// call.  GoSyntax renders jal and jalr calls as CALL; the native mnemonics
+// are accepted too.  The first token must match exactly: a prefix test on
+// "bl" would catch branches on other architectures, and jalr as ret prints
+// RET, which must not be stepped over.
+func isCallInsn(text string) bool {
+	m, _, _ := strings.Cut(text, " ")
+	switch strings.ToLower(m) {
+	case "call", "jal", "jalr":
+		return true
+	}
+	return false
+}

debug/disasm_linux_amd64.go

@@ -9,8 +9,8 @@ import (
 	"fmt"
 	"strings"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/arch"
-	"sourcedock.dev/petrbalvin/gasm-devkit/disasm"
+	"sourcedock.dev/petrbalvin/gasm-sdk/arch"
+	"sourcedock.dev/petrbalvin/gasm-sdk/disasm"
 )
 
 // Disassemble decodes the instruction at the given address in the debuggee's

debug/disasm_linux_arm64.go

@@ -9,8 +9,8 @@ import (
 	"fmt"
 	"strings"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/arch"
-	"sourcedock.dev/petrbalvin/gasm-devkit/disasm"
+	"sourcedock.dev/petrbalvin/gasm-sdk/arch"
+	"sourcedock.dev/petrbalvin/gasm-sdk/disasm"
 )
 
 // Disassemble decodes the instruction at the given address in the debuggee's

debug/disasm_linux_loong64.go

@@ -9,8 +9,8 @@ import (
 	"fmt"
 	"strings"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/arch"
-	"sourcedock.dev/petrbalvin/gasm-devkit/disasm"
+	"sourcedock.dev/petrbalvin/gasm-sdk/arch"
+	"sourcedock.dev/petrbalvin/gasm-sdk/disasm"
 )
 
 // Disassemble decodes the instruction at the given address in the debuggee's

debug/disasm_linux_riscv64.go

@@ -9,8 +9,8 @@ import (
 	"fmt"
 	"strings"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/arch"
-	"sourcedock.dev/petrbalvin/gasm-devkit/disasm"
+	"sourcedock.dev/petrbalvin/gasm-sdk/arch"
+	"sourcedock.dev/petrbalvin/gasm-sdk/disasm"
 )
 
 // Disassemble decodes the instruction at the given address in the debuggee's

debug/display_freebsd_amd64.go

@@ -0,0 +1,136 @@
+// Copyright (c) 2026 Petr Balvín <opensource@petrbalvin.org> (https://petrbalvin.org)
+// SPDX-License-Identifier: BSD-3-Clause
+
+//go:build freebsd && amd64
+
+package debug
+
+import "fmt"
+
+func printRegs(regs *Regs, codeBase, funcOff uint64) {
+	fmt.Printf("  RIP = %#016x  (func+%#x)\n", regs.RIP, regs.RIP-codeBase-funcOff)
+	fmt.Printf("  RSP = %#016x  RBP = %#016x\n", regs.RSP, regs.RBP)
+	fmt.Printf("  RAX = %#016x  RBX = %#016x\n", regs.RAX, regs.RBX)
+	fmt.Printf("  RCX = %#016x  RDX = %#016x\n", regs.RCX, regs.RDX)
+	fmt.Printf("  RSI = %#016x  RDI = %#016x\n", regs.RSI, regs.RDI)
+	fmt.Printf("  R8  = %#016x  R9  = %#016x\n", regs.R8, regs.R9)
+	fmt.Printf("  R10 = %#016x  R11 = %#016x\n", regs.R10, regs.R11)
+	fmt.Printf("  R12 = %#016x  R13 = %#016x\n", regs.R12, regs.R13)
+	fmt.Printf("  R14 = %#016x  R15 = %#016x\n", regs.R14, regs.R15)
+	fmt.Printf("  RFLAGS = %#x  [%s]\n", regs.RFLAGS, decodeRflags(regs.RFLAGS))
+}
+
+func printVectorRegs(v *VectorRegs) {
+	fmt.Println("\n  Vector registers (YMM):")
+	for i := 0; i < 16; i += 2 {
+		fmt.Printf("  YMM%-2d = ", i)
+		printYMM(v.YMM[i][:])
+		fmt.Printf("  YMM%-2d = ", i+1)
+		printYMM(v.YMM[i+1][:])
+		fmt.Println()
+	}
+}
+
+func printYMM(b []byte) {
+	for j := 0; j < 32; j += 4 {
+		v := uint32(b[j]) | uint32(b[j+1])<<8 | uint32(b[j+2])<<16 | uint32(b[j+3])<<24
+		fmt.Printf("%08x ", v)
+	}
+}
+
+func decodeRflags(f uint64) string {
+	var flags string
+	if f&1 != 0 {
+		flags += "CF "
+	}
+	if f&(1<<2) != 0 {
+		flags += "PF "
+	}
+	if f&(1<<4) != 0 {
+		flags += "AF "
+	}
+	if f&(1<<6) != 0 {
+		flags += "ZF "
+	}
+	if f&(1<<7) != 0 {
+		flags += "SF "
+	}
+	if f&(1<<8) != 0 {
+		flags += "TF "
+	}
+	if f&(1<<9) != 0 {
+		flags += "IF "
+	}
+	if f&(1<<10) != 0 {
+		flags += "DF "
+	}
+	if f&(1<<11) != 0 {
+		flags += "OF "
+	}
+	if flags == "" {
+		return "none"
+	}
+	return flags[:len(flags)-1]
+}
+
+// SetReg modifies a register value in the debuggee.
+func (s *Session) SetReg(name string, value uint64) error {
+	regs, err := s.GetRegs()
+	if err != nil {
+		return err
+	}
+	switch name {
+	case "rax", "eax", "ax", "al":
+		regs.RAX = value
+	case "rbx", "ebx", "bx", "bl":
+		regs.RBX = value
+	case "rcx", "ecx", "cx", "cl":
+		regs.RCX = value
+	case "rdx", "edx", "dx", "dl":
+		regs.RDX = value
+	case "rsi", "esi", "si":
+		regs.RSI = value
+	case "rdi", "edi", "di":
+		regs.RDI = value
+	case "rbp", "ebp", "bp":
+		regs.RBP = value
+	case "rsp", "esp", "sp":
+		regs.RSP = value
+	case "r8":
+		regs.R8 = value
+	case "r9":
+		regs.R9 = value
+	case "r10":
+		regs.R10 = value
+	case "r11":
+		regs.R11 = value
+	case "r12":
+		regs.R12 = value
+	case "r13":
+		regs.R13 = value
+	case "r14":
+		regs.R14 = value
+	case "r15":
+		regs.R15 = value
+	case "rip", "eip":
+		regs.RIP = value
+	default:
+		return fmt.Errorf("debug: unknown register %q", name)
+	}
+	return s.SetRegs(&regs)
+}
+
+// archReturnAddr reads the return address of the current frame (amd64
+// ABI0 convention).  A function that contains a CALL (or has a frame) is
+// assembled with the prologue PUSHQ BP; MOVQ SP, BP, so mid-function the
+// word at SP is the saved caller BP, a stack address, and the return
+// address sits further up.  Walk the stack from SP and take the first word
+// that lies in an executable mapping: stack and data words never do, a
+// return address always does.  FreeBSD exposes no mapping list, so the
+// walk degenerates to the raw entry convention, [SP] before any push.
+func archReturnAddr(s *Session, regs *Regs) (uint64, error) {
+	return s.Peek(regs.RSP)
+}
+
+// archSPLabel returns the SP register name for display.
+func archSPLabel() string { return "RSP" }

debug/display_freebsd_arm64.go

@@ -0,0 +1,127 @@
+// Copyright (c) 2026 Petr Balvín <opensource@petrbalvin.org> (https://petrbalvin.org)
+// SPDX-License-Identifier: BSD-3-Clause
+
+//go:build freebsd && arm64
+
+package debug
+
+import (
+	"encoding/binary"
+	"fmt"
+)
+
+func printRegs(regs *Regs, codeBase, funcOff uint64) {
+	fmt.Printf("  PC  = %#016x  (func+%#x)\n", regs.PC, regs.PC-codeBase-funcOff)
+	fmt.Printf("  SP  = %#016x  FP  = %#016x\n", regs.SP, regs.X29)
+	fmt.Printf("  LR  = %#016x\n", regs.X30)
+	fmt.Printf("  X0  = %#016x  X1  = %#016x\n", regs.X0, regs.X1)
+	fmt.Printf("  X2  = %#016x  X3  = %#016x\n", regs.X2, regs.X3)
+	fmt.Printf("  X4  = %#016x  X5  = %#016x\n", regs.X4, regs.X5)
+	fmt.Printf("  X6  = %#016x  X7  = %#016x\n", regs.X6, regs.X7)
+	fmt.Printf("  X8  = %#016x  X9  = %#016x\n", regs.X8, regs.X9)
+	fmt.Printf("  X10 = %#016x  X11 = %#016x\n", regs.X10, regs.X11)
+	fmt.Printf("  X12 = %#016x  X13 = %#016x\n", regs.X12, regs.X13)
+	fmt.Printf("  X14 = %#016x  X15 = %#016x\n", regs.X14, regs.X15)
+	fmt.Printf("  X16 = %#016x  X17 = %#016x\n", regs.X16, regs.X17)
+	fmt.Printf("  X18 = %#016x  X19 = %#016x\n", regs.X18, regs.X19)
+	fmt.Printf("  X20 = %#016x  X21 = %#016x\n", regs.X20, regs.X21)
+	fmt.Printf("  X22 = %#016x  X23 = %#016x\n", regs.X22, regs.X23)
+	fmt.Printf("  X24 = %#016x  X25 = %#016x\n", regs.X24, regs.X25)
+	fmt.Printf("  X26 = %#016x  X27 = %#016x\n", regs.X26, regs.X27)
+	fmt.Printf("  X28 = %#016x  PSTATE = %#x\n", regs.X28, regs.PSTATE)
+}
+
+func printVectorRegs(v *VectorRegs) {
+	fmt.Println("\n  Vector registers (V0-V31):")
+	for i := 0; i < 32; i += 2 {
+		fmt.Printf("  V%-2d  = %016x%016x\n", i, binary.LittleEndian.Uint64(v.V[i][8:16]), binary.LittleEndian.Uint64(v.V[i][0:8]))
+		fmt.Printf("  V%-2d  = %016x%016x\n", i+1, binary.LittleEndian.Uint64(v.V[i+1][8:16]), binary.LittleEndian.Uint64(v.V[i+1][0:8]))
+	}
+}
+
+// SetReg modifies a register value in the debuggee.
+func (s *Session) SetReg(name string, value uint64) error {
+	regs, err := s.GetRegs()
+	if err != nil {
+		return err
+	}
+	switch name {
+	case "x0":
+		regs.X0 = value
+	case "x1":
+		regs.X1 = value
+	case "x2":
+		regs.X2 = value
+	case "x3":
+		regs.X3 = value
+	case "x4":
+		regs.X4 = value
+	case "x5":
+		regs.X5 = value
+	case "x6":
+		regs.X6 = value
+	case "x7":
+		regs.X7 = value
+	case "x8":
+		regs.X8 = value
+	case "x9":
+		regs.X9 = value
+	case "x10":
+		regs.X10 = value
+	case "x11":
+		regs.X11 = value
+	case "x12":
+		regs.X12 = value
+	case "x13":
+		regs.X13 = value
+	case "x14":
+		regs.X14 = value
+	case "x15":
+		regs.X15 = value
+	case "x16":
+		regs.X16 = value
+	case "x17":
+		regs.X17 = value
+	case "x18":
+		regs.X18 = value
+	case "x19":
+		regs.X19 = value
+	case "x20":
+		regs.X20 = value
+	case "x21":
+		regs.X21 = value
+	case "x22":
+		regs.X22 = value
+	case "x23":
+		regs.X23 = value
+	case "x24":
+		regs.X24 = value
+	case "x25":
+		regs.X25 = value
+	case "x26":
+		regs.X26 = value
+	case "x27":
+		regs.X27 = value
+	case "x28":
+		regs.X28 = value
+	case "x29", "fp":
+		regs.X29 = value
+	case "x30", "lr":
+		regs.X30 = value
+	case "sp":
+		regs.SP = value
+	case "pc":
+		regs.PC = value
+	default:
+		return fmt.Errorf("debug: unknown register %q", name)
+	}
+	return s.SetRegs(&regs)
+}
+
+// archReturnAddr reads the return address from LR (arm64 convention).
+func archReturnAddr(s *Session, regs *Regs) (uint64, error) {
+	return regs.X30, nil
+}
+
+// archSPLabel returns the SP register name for display.
+func archSPLabel() string { return "SP" }

debug/display_freebsd_riscv64.go

@@ -0,0 +1,121 @@
+// Copyright (c) 2026 Petr Balvín <opensource@petrbalvin.org> (https://petrbalvin.org)
+// SPDX-License-Identifier: BSD-3-Clause
+
+//go:build freebsd && riscv64
+
+package debug
+
+import "fmt"
+
+func printRegs(regs *Regs, codeBase, funcOff uint64) {
+	fmt.Printf("  PC  = %#016x  (func+%#x)\n", regs.PC, regs.PC-codeBase-funcOff)
+	fmt.Printf("  SP  = %#016x  FP  = %#016x\n", regs.Sp, regs.S0)
+	fmt.Printf("  RA  = %#016x\n", regs.Ra)
+	fmt.Printf("  A0  = %#016x  A1  = %#016x\n", regs.A0, regs.A1)
+	fmt.Printf("  A2  = %#016x  A3  = %#016x\n", regs.A2, regs.A3)
+	fmt.Printf("  A4  = %#016x  A5  = %#016x\n", regs.A4, regs.A5)
+	fmt.Printf("  A6  = %#016x  A7  = %#016x\n", regs.A6, regs.A7)
+	fmt.Printf("  T0  = %#016x  T1  = %#016x\n", regs.T0, regs.T1)
+	fmt.Printf("  T2  = %#016x  T3  = %#016x\n", regs.T2, regs.T3)
+	fmt.Printf("  T4  = %#016x  T5  = %#016x\n", regs.T4, regs.T5)
+	fmt.Printf("  T6  = %#016x\n", regs.T6)
+	fmt.Printf("  S1  = %#016x  S2  = %#016x\n", regs.S1, regs.S2)
+	fmt.Printf("  S3  = %#016x  S4  = %#016x\n", regs.S3, regs.S4)
+	fmt.Printf("  S5  = %#016x  S6  = %#016x\n", regs.S5, regs.S6)
+	fmt.Printf("  S7  = %#016x  S8  = %#016x\n", regs.S7, regs.S8)
+	fmt.Printf("  S9  = %#016x  S10 = %#016x\n", regs.S9, regs.S10)
+	fmt.Printf("  S11 = %#016x\n", regs.S11)
+}
+
+func printVectorRegs(v *VectorRegs) {
+	fmt.Println("\n  FP registers (F0-F31):")
+	for i := 0; i < 32; i += 2 {
+		fmt.Printf("  F%-2d  = %#018x  F%-2d  = %#018x\n", i, v.F[i], i+1, v.F[i+1])
+	}
+	fmt.Printf("  FCSR = %#x\n", v.FCSR)
+}
+
+// SetReg modifies a register value in the debuggee.
+func (s *Session) SetReg(name string, value uint64) error {
+	regs, err := s.GetRegs()
+	if err != nil {
+		return err
+	}
+	switch name {
+	case "pc":
+		regs.PC = value
+	case "ra", "x1":
+		regs.Ra = value
+	case "sp", "x2":
+		regs.Sp = value
+	case "gp", "x3":
+		regs.Gp = value
+	case "tp", "x4":
+		regs.Tp = value
+	case "t0", "x5":
+		regs.T0 = value
+	case "t1", "x6":
+		regs.T1 = value
+	case "t2", "x7":
+		regs.T2 = value
+	case "s0", "fp", "x8":
+		regs.S0 = value
+	case "s1", "x9":
+		regs.S1 = value
+	case "a0", "x10":
+		regs.A0 = value
+	case "a1", "x11":
+		regs.A1 = value
+	case "a2", "x12":
+		regs.A2 = value
+	case "a3", "x13":
+		regs.A3 = value
+	case "a4", "x14":
+		regs.A4 = value
+	case "a5", "x15":
+		regs.A5 = value
+	case "a6", "x16":
+		regs.A6 = value
+	case "a7", "x17":
+		regs.A7 = value
+	case "s2", "x18":
+		regs.S2 = value
+	case "s3", "x19":
+		regs.S3 = value
+	case "s4", "x20":
+		regs.S4 = value
+	case "s5", "x21":
+		regs.S5 = value
+	case "s6", "x22":
+		regs.S6 = value
+	case "s7", "x23":
+		regs.S7 = value
+	case "s8", "x24":
+		regs.S8 = value
+	case "s9", "x25":
+		regs.S9 = value
+	case "s10", "x26":
+		regs.S10 = value
+	case "s11", "x27":
+		regs.S11 = value
+	case "t3", "x28":
+		regs.T3 = value
+	case "t4", "x29":
+		regs.T4 = value
+	case "t5", "x30":
+		regs.T5 = value
+	case "t6", "x31":
+		regs.T6 = value
+	default:
+		return fmt.Errorf("debug: unknown register %q", name)
+	}
+	return s.SetRegs(&regs)
+}
+
+// archReturnAddr reads the return address from RA (riscv64 convention).
+func archReturnAddr(s *Session, regs *Regs) (uint64, error) {
+	return regs.Ra, nil
+}
+
+// archSPLabel returns the SP register name for display.
+func archSPLabel() string { return "SP" }

debug/ptrace_extra_test.go

@@ -17,8 +17,8 @@ import (
 	"time"
 	"unsafe"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/asm"
-	"sourcedock.dev/petrbalvin/gasm-devkit/verify"
+	"sourcedock.dev/petrbalvin/gasm-sdk/asm"
+	"sourcedock.dev/petrbalvin/gasm-sdk/verify"
 )
 
 // Integration tests beyond the basic entry breakpoint: hardware watchpoints,

debug/ptrace_freebsd.go

@@ -0,0 +1,288 @@
+// Copyright (c) 2026 Petr Balvín <opensource@petrbalvin.org> (https://petrbalvin.org)
+// SPDX-License-Identifier: BSD-3-Clause
+
+//go:build freebsd && (amd64 || arm64 || riscv64)
+
+package debug
+
+import (
+	"fmt"
+	"os"
+	"os/exec"
+	"path/filepath"
+	"runtime"
+	"strings"
+	"syscall"
+	"time"
+
+	"golang.org/x/sys/unix"
+)
+
+// Session is a ptrace debugging session controlling one debuggee process.
+// The FreeBSD implementation sits behind the same surface as the Linux one:
+// PT_TRACE_ME from the debuggee, PT_CONTINUE/PT_STEP from the tracer, and
+// tracee memory through PT_IO (FreeBSD has no /proc/pid/mem to fall back
+// on, so PT_IO is the only supported route).
+type Session struct {
+	pid      int
+	cmd      *exec.Cmd
+	stopped  bool
+	exited   bool
+	codeBase uint64   // base address of the JIT code in the debuggee
+	tmpDir   string   // scratch directory of the session, removed on Kill
+	wpSlots  [16]bool // hardware watchpoint slots in use (DR0-DR3, arm64 dbw 0-15)
+	// lastSignal holds the signal of the most recent stop when that stop
+	// was a genuine signal-delivery-stop the caller must see (a fault such
+	// as SIGSEGV, SIGBUS, SIGFPE or SIGILL); 0 for breakpoint traps,
+	// single-steps, SIGSTOP and suppressed runtime signals.
+	lastSignal syscall.Signal
+}
+
+// Launch starts the debuggee subprocess (gasm debug --target ...) and
+// attaches to it via ptrace.
+func Launch(gasmBin, asmPath, funcName string, args []byte) (*Session, error) {
+	sess, _, err := LaunchWithBuffers(gasmBin, asmPath, funcName, args, "")
+	return sess, err
+}
+
+// LaunchWithBuffers is like Launch but also allocates buffers in the debuggee.
+//
+// It pins the calling goroutine to its OS thread and leaves it pinned: the
+// debuggee's PT_TRACE_ME binds the tracer relation to the forking thread,
+// and every ptrace request on the session must come from that same thread.
+// All Session methods must therefore be called from the goroutine that
+// launched the session (the REPL and coverage loops do exactly that).
+func LaunchWithBuffers(gasmBin, asmPath, funcName string, args []byte, bufSpec string) (*Session, []uint64, error) {
+	runtime.LockOSThread() // ptrace requests must stay on the forking thread
+	self, err := os.Executable()
+	if err != nil {
+		return nil, nil, fmt.Errorf("debug: cannot find gasm binary: %w", err)
+	}
+	if gasmBin != "" {
+		self = gasmBin
+	}
+
+	tmpDir, err := os.MkdirTemp("", "gasm-debug-*")
+	if err != nil {
+		return nil, nil, fmt.Errorf("debug: tempdir: %w", err)
+	}
+	argsFile := filepath.Join(tmpDir, "args.bin")
+	if err := os.WriteFile(argsFile, args, 0o644); err != nil {
+		os.RemoveAll(tmpDir)
+		return nil, nil, fmt.Errorf("debug: write args: %w", err)
+	}
+
+	if bufSpec != "" {
+		if err := os.WriteFile(filepath.Join(tmpDir, "bufspec"), []byte(bufSpec), 0o644); err != nil {
+			os.RemoveAll(tmpDir)
+			return nil, nil, fmt.Errorf("debug: write bufspec: %w", err)
+		}
+	}
+
+	cmd := exec.Command(self, "debug", "--func", funcName, "--args", argsFile, asmPath)
+	cmd.Env = append(os.Environ(), "GASM_DEBUG_TARGET=1", "GASM_DEBUG_TMP="+tmpDir)
+	cmd.Stdout = nil
+	cmd.Stderr = os.Stderr
+	cmd.SysProcAttr = &syscall.SysProcAttr{}
+
+	if err := cmd.Start(); err != nil {
+		os.RemoveAll(tmpDir)
+		return nil, nil, fmt.Errorf("debug: start debuggee: %w", err)
+	}
+
+	s := &Session{pid: cmd.Process.Pid, cmd: cmd, tmpDir: tmpDir}
+
+	readyFile := filepath.Join(tmpDir, "ready")
+	for range 500 {
+		if _, err := os.Stat(readyFile); err == nil {
+			break
+		}
+		time.Sleep(5 * time.Millisecond)
+	}
+
+	// The debuggee parks itself with SIGSTOP once the JIT code is mapped.
+	// A Go tracee also reports SIGURG preemption as signal-delivery-stops,
+	// so the wait loops until a stop the debugger cares about instead of
+	// assuming the first event is the SIGSTOP.
+	if _, err := s.waitStopped(); err != nil {
+		cmd.Process.Kill()
+		os.RemoveAll(tmpDir)
+		return nil, nil, fmt.Errorf("debug: wait for debuggee: %w", err)
+	}
+	s.stopped = true
+
+	// The debuggee reports its JIT mapping in the codebase file; that is
+	// the supported path on FreeBSD, where no /proc/pid/maps exists to
+	// scan for the RWX region as a fallback.
+	if data, err := os.ReadFile(filepath.Join(tmpDir, "codebase")); err == nil {
+		fmt.Sscanf(string(data), "%d", &s.codeBase)
+	}
+
+	var bufAddrs []uint64
+	if bufSpec != "" {
+		addrFile := filepath.Join(tmpDir, "bufaddrs")
+		if data, err := os.ReadFile(addrFile); err == nil {
+			for line := range strings.SplitSeq(strings.TrimSpace(string(data)), "\n") {
+				var addr uint64
+				if _, err := fmt.Sscanf(line, "%d", &addr); err == nil {
+					bufAddrs = append(bufAddrs, addr)
+				}
+			}
+		}
+	}
+
+	return s, bufAddrs, nil
+}
+
+// waitStopped consumes ptrace-stop events until one the debugger cares
+// about arrives: SIGTRAP (a breakpoint or a completed single-step), the
+// debuggee's own SIGSTOP, or a genuine signal-delivery-stop.  A Go tracee's
+// runtime raises SIGURG for asynchronous preemption, and every signal on a
+// traced thread surfaces as a signal-delivery-stop, so SIGURG is suppressed
+// and the tracee resumed without it.  Every other signal (SIGSEGV, SIGBUS,
+// SIGFPE, SIGILL, ...) is returned to the caller: resuming with signal 0
+// would restart the faulting instruction and fault forever, so a faulting
+// kernel must surface as a stop the caller reports.
+func (s *Session) waitStopped() (syscall.Signal, error) {
+	for {
+		var ws syscall.WaitStatus
+		if _, err := syscall.Wait4(s.pid, &ws, syscall.WUNTRACED, nil); err != nil {
+			return 0, err
+		}
+		if ws.Exited() {
+			s.exited = true
+			return 0, fmt.Errorf("debuggee exited with status %d", ws.ExitStatus())
+		}
+		if ws.Signaled() {
+			s.exited = true
+			return 0, fmt.Errorf("debuggee killed by signal %v", ws.Signal())
+		}
+		switch sig := ws.StopSignal(); sig {
+		case syscall.SIGTRAP, syscall.SIGSTOP:
+			s.stopped = true
+			s.lastSignal = 0
+			return sig, nil
+		case syscall.SIGURG:
+			// Go runtime asynchronous preemption: resume the tracee
+			// without delivering the signal.
+			s.lastSignal = 0
+			if err := unix.PtraceCont(s.pid, 0); err != nil {
+				return 0, fmt.Errorf("debug: PT_CONTINUE: %w", err)
+			}
+		default:
+			// A genuine signal-delivery-stop.  Report it; the caller
+			// decides how to proceed.
+			s.stopped = true
+			s.lastSignal = sig
+			return sig, nil
+		}
+	}
+}
+
+// LastSignal returns the signal of the most recent stop when that stop was
+// a genuine signal-delivery-stop (a fault such as SIGSEGV, SIGFPE, SIGILL
+// or SIGBUS), and 0 for breakpoint traps, single-steps, SIGSTOP and
+// suppressed runtime signals.
+func (s *Session) LastSignal() syscall.Signal { return s.lastSignal }
+
+// Peek reads a word (8 bytes) from the debuggee's memory at addr, through
+// PT_IO with PIOD_READ_D.
+func (s *Session) Peek(addr uint64) (uint64, error) {
+	var buf [8]byte
+	if _, err := unix.PtraceIO(unix.PIOD_READ_D, s.pid, uintptr(addr), buf[:], len(buf)); err != nil {
+		return 0, fmt.Errorf("debug: read mem %#x: %w", addr, err)
+	}
+	return uint64(buf[0]) | uint64(buf[1])<<8 | uint64(buf[2])<<16 | uint64(buf[3])<<24 |
+		uint64(buf[4])<<32 | uint64(buf[5])<<40 | uint64(buf[6])<<48 | uint64(buf[7])<<56, nil
+}
+
+// Poke writes a word (8 bytes) to the debuggee's memory at addr, through
+// PT_IO with PIOD_WRITE_D.
+func (s *Session) Poke(addr, val uint64) error {
+	buf := []byte{byte(val), byte(val >> 8), byte(val >> 16), byte(val >> 24),
+		byte(val >> 32), byte(val >> 40), byte(val >> 48), byte(val >> 56)}
+	if _, err := unix.PtraceIO(unix.PIOD_WRITE_D, s.pid, uintptr(addr), buf, len(buf)); err != nil {
+		return fmt.Errorf("debug: write mem %#x: %w", addr, err)
+	}
+	return nil
+}
+
+// ReadMemory reads len bytes from the debuggee's memory at addr in one
+// PT_IO request, the shape the request is built for.
+func (s *Session) ReadMemory(addr uint64, length int) ([]byte, error) {
+	out := make([]byte, length)
+	n, err := unix.PtraceIO(unix.PIOD_READ_D, s.pid, uintptr(addr), out, length)
+	return out[:n], err
+}
+
+// WriteMemory writes bytes to the debuggee's memory at addr in one PT_IO
+// request.
+func (s *Session) WriteMemory(addr uint64, data []byte) error {
+	_, err := unix.PtraceIO(unix.PIOD_WRITE_D, s.pid, uintptr(addr), data, len(data))
+	return err
+}
+
+// Step executes a single instruction in the debuggee.
+func (s *Session) Step() error {
+	if s.exited {
+		return fmt.Errorf("debug: debuggee has exited")
+	}
+	if err := unix.PtraceSingleStep(s.pid); err != nil {
+		return fmt.Errorf("debug: PT_STEP: %w", err)
+	}
+	_, err := s.waitStopped()
+	return err
+}
+
+// Continue resumes execution until the next breakpoint or exit.
+func (s *Session) Continue() error {
+	if s.exited {
+		return fmt.Errorf("debug: debuggee has exited")
+	}
+	if err := unix.PtraceCont(s.pid, 0); err != nil {
+		return fmt.Errorf("debug: PT_CONTINUE: %w", err)
+	}
+	_, err := s.waitStopped()
+	return err
+}
+
+// Exited returns true if the debuggee has terminated.
+func (s *Session) Exited() bool { return s.exited }
+
+// Pid returns the debuggee's process ID.
+func (s *Session) Pid() int { return s.pid }
+
+// CodeBase returns the base address of the JIT code in the debuggee.
+func (s *Session) CodeBase() uint64 { return s.codeBase }
+
+// Kill terminates the debuggee and removes the session's scratch
+// directory, so a successful session leaves no gasm-debug-* debris behind.
+func (s *Session) Kill() {
+	if !s.exited {
+		syscall.Kill(s.pid, syscall.SIGKILL)
+		syscall.Wait4(s.pid, nil, 0, nil)
+		s.exited = true
+	}
+	if s.cmd != nil && s.cmd.Process != nil {
+		s.cmd.Wait()
+	}
+	if s.tmpDir != "" {
+		os.RemoveAll(s.tmpDir)
+		s.tmpDir = ""
+	}
+}
+
+// execRange is one executable mapping of the debuggee.
+type execRange struct {
+	lo, hi uint64
+}
+
+// execRanges is a stub on FreeBSD: there is no /proc/pid/maps to parse,
+// and procfs(5) is not guaranteed to be mounted.  The callers degrade
+// gracefully: archReturnAddr falls back to the raw stack convention and
+// the mapping scan is skipped.
+func execRanges(pid int) []execRange { return nil }
+
+// findRWXMapping is a stub on FreeBSD for the same reason: the codebase
+// handshake file is the supported way the JIT region is located.
+func findRWXMapping(pid int) uint64 { return 0 }

debug/ptrace_freebsd_amd64.go

@@ -0,0 +1,160 @@
+// Copyright (c) 2026 Petr Balvín <opensource@petrbalvin.org> (https://petrbalvin.org)
+// SPDX-License-Identifier: BSD-3-Clause
+
+//go:build freebsd && amd64
+
+package debug
+
+import (
+	"encoding/binary"
+	"fmt"
+	"unsafe"
+
+	"golang.org/x/sys/unix"
+)
+
+// GetRegs reads the general-purpose registers of the stopped debuggee and
+// converts the FreeBSD struct reg into the portable layout.
+func (s *Session) GetRegs() (Regs, error) {
+	var ur unix.Reg
+	if err := unix.PtraceGetRegs(s.pid, &ur); err != nil {
+		return Regs{}, fmt.Errorf("debug: PT_GETREGS: %w", err)
+	}
+	return Regs{
+		R15:    uint64(ur.R15),
+		R14:    uint64(ur.R14),
+		R13:    uint64(ur.R13),
+		R12:    uint64(ur.R12),
+		R11:    uint64(ur.R11),
+		R10:    uint64(ur.R10),
+		R9:     uint64(ur.R9),
+		R8:     uint64(ur.R8),
+		RDI:    uint64(ur.Rdi),
+		RSI:    uint64(ur.Rsi),
+		RBP:    uint64(ur.Rbp),
+		RBX:    uint64(ur.Rbx),
+		RDX:    uint64(ur.Rdx),
+		RCX:    uint64(ur.Rcx),
+		RAX:    uint64(ur.Rax),
+		RIP:    uint64(ur.Rip),
+		CS:     uint64(ur.Cs),
+		RFLAGS: uint64(ur.Rflags),
+		RSP:    uint64(ur.Rsp),
+		SS:     uint64(ur.Ss),
+		FS:     uint64(ur.Fs),
+		GS:     uint64(ur.Gs),
+		DS:     uint64(ur.Ds),
+		ES:     uint64(ur.Es),
+	}, nil
+}
+
+// SetRegs writes the general-purpose registers of the stopped debuggee.
+func (s *Session) SetRegs(regs *Regs) error {
+	// Read-modify-write keeps the fields FreeBSD owns (trapno, err) intact.
+	var ur unix.Reg
+	if err := unix.PtraceGetRegs(s.pid, &ur); err != nil {
+		return fmt.Errorf("debug: PT_GETREGS: %w", err)
+	}
+	ur.R15 = int64(regs.R15)
+	ur.R14 = int64(regs.R14)
+	ur.R13 = int64(regs.R13)
+	ur.R12 = int64(regs.R12)
+	ur.R11 = int64(regs.R11)
+	ur.R10 = int64(regs.R10)
+	ur.R9 = int64(regs.R9)
+	ur.R8 = int64(regs.R8)
+	ur.Rdi = int64(regs.RDI)
+	ur.Rsi = int64(regs.RSI)
+	ur.Rbp = int64(regs.RBP)
+	ur.Rbx = int64(regs.RBX)
+	ur.Rdx = int64(regs.RDX)
+	ur.Rcx = int64(regs.RCX)
+	ur.Rax = int64(regs.RAX)
+	ur.Rip = int64(regs.RIP)
+	ur.Cs = int64(regs.CS)
+	ur.Rflags = int64(regs.RFLAGS)
+	ur.Rsp = int64(regs.RSP)
+	ur.Ss = int64(regs.SS)
+	return unix.PtraceSetRegs(s.pid, &ur)
+}
+
+// FPRegs holds the x87 FPU and SSE (XMM) register state, the FXSAVE image
+// the FreeBSD struct fpreg mirrors: XMM0-15 at the same offsets.
+type FPRegs struct {
+	XMM [16][16]byte // XMM0-15
+}
+
+// GetFPRegs retrieves the FPU/SSE register state via PT_GETFPREGS.  The
+// FreeBSD struct fpreg mirrors the FXSAVE image: the x87 environment and
+// stack in Env/Acc, XMM0-15 in Xacc.
+func (s *Session) GetFPRegs() (FPRegs, error) {
+	var fp FPRegs
+	var fr unix.FpReg
+	if err := unix.PtraceGetFpRegs(s.pid, &fr); err != nil {
+		return fp, fmt.Errorf("debug: PT_GETFPREGS: %w", err)
+	}
+	for i := range 16 {
+		copy(fp.XMM[i][:], fr.Xacc[i][:])
+	}
+	return fp, nil
+}
+
+// VectorRegs holds the YMM register state.
+type VectorRegs struct {
+	YMM [16][32]byte // YMM0-15 (full 256-bit values)
+}
+
+// The XSAVE area the PT_GETXSTATE request returns follows the architectural
+// layout (Intel SDM vol 1, "XSAVE"): the 512-byte legacy FXSAVE image (x87
+// state in 0-159, XMM0-15 in 160-511), then the 64-byte xsave header whose
+// first 8 bytes are xstate_bv, then one component per set feature bit, each
+// 64-byte aligned.  The YMM high halves are the first extended component,
+// at offset 576; XFEATURE_STATE_BIT_AVX is bit 2 of xstate_bv.
+const (
+	xsaveXMMOffset    = 160
+	xsaveHeaderOffset = 512
+	xsaveBVOffset     = xsaveHeaderOffset
+	ymmOffset         = xsaveHeaderOffset + 64 // 576
+	ymmSize           = 256                    // 16 registers, 16 bytes each
+	xfeatureMaskYMM   = 1 << 2
+	xstateMaxBuffer   = 4096 // PT_GETXSTATE_INFO bounds the size far below this
+)
+
+// GetVectorRegs retrieves the YMM registers via PT_GETXSTATE.  The low
+// (XMM) halves always come from the legacy image; the high halves are
+// copied only when xstate_bv reports the AVX state, and read as zero
+// otherwise.  When the request fails the FP image still provides correct
+// XMM halves, so that is the fallback.
+func (s *Session) GetVectorRegs() (VectorRegs, error) {
+	var v VectorRegs
+	buf := make([]byte, xstateMaxBuffer)
+	n, _, errno := unix.Syscall6(
+		unix.SYS_PTRACE,
+		uintptr(unix.PT_GETXSTATE),
+		uintptr(s.pid),
+		0,
+		uintptr(unsafe.Pointer(&buf[0])),
+		0, 0,
+	)
+	if errno != 0 {
+		fp, err := s.GetFPRegs()
+		if err != nil {
+			return v, err
+		}
+		for i := range 16 {
+			copy(v.YMM[i][:16], fp.XMM[i][:])
+		}
+		return v, nil
+	}
+	for i := range 16 {
+		copy(v.YMM[i][:16], buf[xsaveXMMOffset+16*i:xsaveXMMOffset+16*i+16])
+	}
+	if int(n) >= ymmOffset+ymmSize {
+		if binary.LittleEndian.Uint64(buf[xsaveBVOffset:xsaveBVOffset+8])&xfeatureMaskYMM != 0 {
+			for i := range 16 {
+				copy(v.YMM[i][16:], buf[ymmOffset+16*i:ymmOffset+16*i+16])
+			}
+		}
+	}
+	return v, nil
+}

debug/ptrace_freebsd_arm64.go

@@ -0,0 +1,114 @@
+// Copyright (c) 2026 Petr Balvín <opensource@petrbalvin.org> (https://petrbalvin.org)
+// SPDX-License-Identifier: BSD-3-Clause
+
+//go:build freebsd && arm64
+
+package debug
+
+import (
+	"fmt"
+
+	"golang.org/x/sys/unix"
+)
+
+// GetRegs reads the general-purpose registers of the stopped debuggee and
+// converts the FreeBSD struct reg (x[30], lr, sp, elr, spsr) into the
+// portable layout.
+func (s *Session) GetRegs() (Regs, error) {
+	var ur unix.Reg
+	if err := unix.PtraceGetRegs(s.pid, &ur); err != nil {
+		return Regs{}, fmt.Errorf("debug: PT_GETREGS: %w", err)
+	}
+	return Regs{
+		X0:     ur.X[0],
+		X1:     ur.X[1],
+		X2:     ur.X[2],
+		X3:     ur.X[3],
+		X4:     ur.X[4],
+		X5:     ur.X[5],
+		X6:     ur.X[6],
+		X7:     ur.X[7],
+		X8:     ur.X[8],
+		X9:     ur.X[9],
+		X10:    ur.X[10],
+		X11:    ur.X[11],
+		X12:    ur.X[12],
+		X13:    ur.X[13],
+		X14:    ur.X[14],
+		X15:    ur.X[15],
+		X16:    ur.X[16],
+		X17:    ur.X[17],
+		X18:    ur.X[18],
+		X19:    ur.X[19],
+		X20:    ur.X[20],
+		X21:    ur.X[21],
+		X22:    ur.X[22],
+		X23:    ur.X[23],
+		X24:    ur.X[24],
+		X25:    ur.X[25],
+		X26:    ur.X[26],
+		X27:    ur.X[27],
+		X28:    ur.X[28],
+		X29:    ur.X[29],
+		X30:    ur.Lr,
+		SP:     ur.Sp,
+		PC:     ur.Elr,
+		PSTATE: uint64(ur.Spsr),
+	}, nil
+}
+
+// SetRegs writes the general-purpose registers of the stopped debuggee.
+func (s *Session) SetRegs(regs *Regs) error {
+	var ur unix.Reg
+	ur.X = [30]uint64{
+		regs.X0, regs.X1, regs.X2, regs.X3, regs.X4, regs.X5, regs.X6,
+		regs.X7, regs.X8, regs.X9, regs.X10, regs.X11, regs.X12, regs.X13,
+		regs.X14, regs.X15, regs.X16, regs.X17, regs.X18, regs.X19, regs.X20,
+		regs.X21, regs.X22, regs.X23, regs.X24, regs.X25, regs.X26, regs.X27,
+		regs.X28, regs.X29,
+	}
+	ur.Lr = regs.X30
+	ur.Sp = regs.SP
+	ur.Elr = regs.PC
+	ur.Spsr = uint32(regs.PSTATE)
+	return unix.PtraceSetRegs(s.pid, &ur)
+}
+
+// FPRegs holds the arm64 FP/NEON register state: the 32 128-bit V
+// registers, then FPSR and FPCR (the user_fpsimd shape).
+type FPRegs struct {
+	V    [32][16]byte // V0-V31 (128-bit NEON/FP registers)
+	FPSR uint32
+	FPCR uint32
+}
+
+// GetFPRegs retrieves the FP/NEON register state via PT_GETFPREGS.  The
+// FreeBSD struct fpreg holds the 32 128-bit V registers followed by FPSR
+// and FPCR, the user_fpsimd shape.
+func (s *Session) GetFPRegs() (FPRegs, error) {
+	var fp FPRegs
+	var fr unix.FpReg
+	if err := unix.PtraceGetFpRegs(s.pid, &fr); err != nil {
+		return fp, fmt.Errorf("debug: PT_GETFPREGS: %w", err)
+	}
+	for i := range 32 {
+		copy(fp.V[i][:], fr.Q[i][:])
+	}
+	return fp, nil
+}
+
+// VectorRegs holds the full SIMD register state.
+type VectorRegs struct {
+	V [32][16]byte // V0-V31 (128-bit)
+}
+
+// GetVectorRegs retrieves the SIMD registers.
+func (s *Session) GetVectorRegs() (VectorRegs, error) {
+	var v VectorRegs
+	fp, err := s.GetFPRegs()
+	if err != nil {
+		return v, err
+	}
+	copy(v.V[:][:], fp.V[:][:])
+	return v, nil
+}

debug/ptrace_freebsd_riscv64.go

@@ -0,0 +1,115 @@
+// Copyright (c) 2026 Petr Balvín <opensource@petrbalvin.org> (https://petrbalvin.org)
+// SPDX-License-Identifier: BSD-3-Clause
+
+//go:build freebsd && riscv64
+
+package debug
+
+import (
+	"fmt"
+
+	"golang.org/x/sys/unix"
+)
+
+// GetRegs reads the general-purpose registers of the stopped debuggee and
+// converts the FreeBSD struct reg into the portable layout.  Sstatus rides
+// the kernel's struct but the portable surface carries the GPRs and PC.
+func (s *Session) GetRegs() (Regs, error) {
+	var ur unix.Reg
+	if err := unix.PtraceGetRegs(s.pid, &ur); err != nil {
+		return Regs{}, fmt.Errorf("debug: PT_GETREGS: %w", err)
+	}
+	return Regs{
+		PC:  ur.Sepc,
+		Ra:  ur.Ra,
+		Sp:  ur.Sp,
+		Gp:  ur.Gp,
+		Tp:  ur.Tp,
+		T0:  ur.T[0],
+		T1:  ur.T[1],
+		T2:  ur.T[2],
+		S0:  ur.S[0],
+		S1:  ur.S[1],
+		A0:  ur.A[0],
+		A1:  ur.A[1],
+		A2:  ur.A[2],
+		A3:  ur.A[3],
+		A4:  ur.A[4],
+		A5:  ur.A[5],
+		A6:  ur.A[6],
+		A7:  ur.A[7],
+		S2:  ur.S[2],
+		S3:  ur.S[3],
+		S4:  ur.S[4],
+		S5:  ur.S[5],
+		S6:  ur.S[6],
+		S7:  ur.S[7],
+		S8:  ur.S[8],
+		S9:  ur.S[9],
+		S10: ur.S[10],
+		S11: ur.S[11],
+		T3:  ur.T[3],
+		T4:  ur.T[4],
+		T5:  ur.T[5],
+		T6:  ur.T[6],
+	}, nil
+}
+
+// SetRegs writes the general-purpose registers of the stopped debuggee.
+// Read-modify-write keeps sstatus, which the kernel owns, intact.
+func (s *Session) SetRegs(regs *Regs) error {
+	var ur unix.Reg
+	if err := unix.PtraceGetRegs(s.pid, &ur); err != nil {
+		return fmt.Errorf("debug: PT_GETREGS: %w", err)
+	}
+	ur.Sepc = regs.PC
+	ur.Ra = regs.Ra
+	ur.Sp = regs.Sp
+	ur.Gp = regs.Gp
+	ur.Tp = regs.Tp
+	ur.T = [7]uint64{regs.T0, regs.T1, regs.T2, regs.T3, regs.T4, regs.T5, regs.T6}
+	ur.S = [12]uint64{regs.S0, regs.S1, regs.S2, regs.S3, regs.S4, regs.S5,
+		regs.S6, regs.S7, regs.S8, regs.S9, regs.S10, regs.S11}
+	ur.A = [8]uint64{regs.A0, regs.A1, regs.A2, regs.A3, regs.A4, regs.A5, regs.A6, regs.A7}
+	return unix.PtraceSetRegs(s.pid, &ur)
+}
+
+// FPRegs holds the RISC-V FP register state (32 64-bit FP registers plus
+// fcsr).
+type FPRegs struct {
+	F    [32]uint64 // F0-F31 (64-bit FP registers)
+	FCSR uint32
+}
+
+// GetFPRegs retrieves the FP register state via PT_GETFPREGS.  The FreeBSD
+// struct fpreg carries each 64-bit FP register in a 128-bit slot (fp_x is
+// the flat [64]-word area the x/sys type renders as [32][2]); the low word
+// holds the register, and FCSR rides the tail.
+func (s *Session) GetFPRegs() (FPRegs, error) {
+	var fp FPRegs
+	var fr unix.FpReg
+	if err := unix.PtraceGetFpRegs(s.pid, &fr); err != nil {
+		return fp, fmt.Errorf("debug: PT_GETFPREGS: %w", err)
+	}
+	for i := range 32 {
+		fp.F[i] = fr.X[i][0]
+	}
+	fp.FCSR = uint32(fr.Fcsr)
+	return fp, nil
+}
+
+// VectorRegs holds the FP register state shown by the regs command
+// (riscv64 has 32 64-bit FP registers and fcsr).
+type VectorRegs struct {
+	F    [32]uint64
+	FCSR uint32
+}
+
+// GetVectorRegs retrieves the FP registers.
+func (s *Session) GetVectorRegs() (VectorRegs, error) {
+	fp, err := s.GetFPRegs()
+	if err != nil {
+		return VectorRegs{}, err
+	}
+	return VectorRegs{F: fp.F, FCSR: fp.FCSR}, nil
+}

debug/ptrace_integration_freebsd_test.go

@@ -0,0 +1,91 @@
+// Copyright (c) 2026 Petr Balvín <opensource@petrbalvin.org> (https://petrbalvin.org)
+// SPDX-License-Identifier: BSD-3-Clause
+
+//go:build freebsd && amd64
+
+package debug
+
+import (
+	"os/exec"
+	"path/filepath"
+	"runtime"
+	"testing"
+
+	"sourcedock.dev/petrbalvin/gasm-sdk/verify"
+)
+
+// TestLaunchAndBreakpoint is the FreeBSD twin of the Linux integration
+// test: it drives the whole launch, breakpoint, trap and register-rewind
+// flow end to end.  It needs a real FreeBSD kernel (ptrace does not work
+// under emulation), so it only runs where it can.
+func TestLaunchAndBreakpoint(t *testing.T) {
+	if runtime.GOARCH != "amd64" {
+		t.Skip("runs only on amd64 hosts")
+	}
+	// The tracer is the OS thread that forked the debuggee (PT_TRACE_ME
+	// binds the relation to that thread); every ptrace request must come
+	// from the same thread, so pin the test goroutine to one thread.
+	runtime.LockOSThread()
+	defer runtime.UnlockOSThread()
+
+	bin := filepath.Join(t.TempDir(), "gasm")
+	out, err := exec.Command("go", "build", "-o", bin, "sourcedock.dev/petrbalvin/gasm-sdk/cmd/gasm").CombinedOutput()
+	if err != nil {
+		t.Fatalf("build gasm: %v: %s", err, out)
+	}
+
+	const kernelPath = "../testdata/verify/basic_amd64.s"
+	k, err := verify.Load(kernelPath)
+	if err != nil {
+		t.Fatalf("Load: %v", err)
+	}
+	t.Cleanup(k.Close)
+	fl, err := k.Func("wideCopy")
+	if err != nil {
+		t.Fatalf("Func: %v", err)
+	}
+
+	sess, err := Launch(bin, kernelPath, "wideCopy", make([]byte, fl.Args))
+	if err != nil {
+		t.Fatalf("Launch: %v", err)
+	}
+	t.Cleanup(sess.Kill)
+
+	bm := NewBreakpoints(sess)
+	entry := sess.CodeBase() + uint64(fl.Offset)
+	if _, err := bm.Set(entry, "entry"); err != nil {
+		t.Fatalf("Set: %v", err)
+	}
+
+	// The INT3 must be visible in the debuggee's memory.
+	word, err := sess.Peek(entry)
+	if err != nil {
+		t.Fatalf("Peek: %v", err)
+	}
+	if b := word & 0xFF; b != 0xCC {
+		t.Fatalf("int3 not patched: first byte %#02x at %#x", b, entry)
+	}
+
+	// The debuggee raises a second SIGSTOP after the launch barrier (the
+	// child's RunTarget marks its entry), so like the REPL and the cover
+	// mode the test keeps resuming until the breakpoint trap arrives.
+	for range 10 {
+		if err := sess.Continue(); err != nil {
+			t.Fatalf("Continue: %v", err)
+		}
+		if sess.Exited() {
+			t.Fatal("debuggee exited instead of trapping on the breakpoint")
+		}
+		regs, err := sess.GetRegs()
+		if err != nil {
+			t.Fatalf("GetRegs: %v", err)
+		}
+		if bp := bm.HandleTrap(&regs); bp != nil {
+			if bp.Addr != entry {
+				t.Fatalf("trap at %#x, want %#x", bp.Addr, entry)
+			}
+			return // trap on the entry breakpoint: the whole flow works
+		}
+	}
+	t.Fatal("no breakpoint trap after 10 resumes")
+}

debug/ptrace_integration_test.go

@@ -14,7 +14,7 @@ import (
 	"strings"
 	"testing"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/verify"
+	"sourcedock.dev/petrbalvin/gasm-sdk/verify"
 )
 
 // buildGasm produces the gasm binary the debugger spawns as its debuggee.
@@ -24,7 +24,7 @@ func buildGasm(t *testing.T) string {
 		return p
 	}
 	bin := filepath.Join(t.TempDir(), "gasm")
-	cmd := exec.Command("go", "build", "-o", bin, "sourcedock.dev/petrbalvin/gasm-devkit/cmd/gasm")
+	cmd := exec.Command("go", "build", "-o", bin, "sourcedock.dev/petrbalvin/gasm-sdk/cmd/gasm")
 	out, err := cmd.CombinedOutput()
 	if err != nil {
 		t.Fatalf("build gasm: %v: %s", err, out)

debug/regs_freebsd_amd64.go

@@ -0,0 +1,98 @@
+// Copyright (c) 2026 Petr Balvín <opensource@petrbalvin.org> (https://petrbalvin.org)
+// SPDX-License-Identifier: BSD-3-Clause
+
+//go:build freebsd && amd64
+
+package debug
+
+// Regs holds the full general-purpose register set of a traced process
+// (the FreeBSD amd64 struct reg layout, sys/x86/include/reg.h).  FreeBSD
+// reports segment selectors (FS/GS/ES/DS), not the bases the Linux ptrace
+// surface carries, and has no ORIG_RAX slot.
+type Regs struct {
+	R15    uint64
+	R14    uint64
+	R13    uint64
+	R12    uint64
+	RBP    uint64
+	RBX    uint64
+	R11    uint64
+	R10    uint64
+	R9     uint64
+	R8     uint64
+	RAX    uint64
+	RCX    uint64
+	RDX    uint64
+	RSI    uint64
+	RDI    uint64
+	RIP    uint64
+	CS     uint64
+	RFLAGS uint64
+	RSP    uint64
+	SS     uint64
+	FS     uint64
+	GS     uint64
+	DS     uint64
+	ES     uint64
+}
+
+// GetPC returns the program counter.
+func (r *Regs) GetPC() uint64 { return r.RIP }
+
+// SetPC sets the program counter.
+func (r *Regs) SetPC(pc uint64) { r.RIP = pc }
+
+// GetSP returns the stack pointer.
+func (r *Regs) GetSP() uint64 { return r.RSP }
+
+// RegValue returns the value of the named register, or false if unknown.
+func (r *Regs) RegValue(name string) (uint64, bool) {
+	switch name {
+	case "rax", "eax", "ax", "al":
+		return r.RAX, true
+	case "rbx", "ebx", "bx", "bl":
+		return r.RBX, true
+	case "rcx", "ecx", "cx", "cl":
+		return r.RCX, true
+	case "rdx", "edx", "dx", "dl":
+		return r.RDX, true
+	case "rsi", "esi", "si":
+		return r.RSI, true
+	case "rdi", "edi", "di":
+		return r.RDI, true
+	case "rbp", "ebp", "bp":
+		return r.RBP, true
+	case "rsp", "esp", "sp":
+		return r.RSP, true
+	case "r8":
+		return r.R8, true
+	case "r9":
+		return r.R9, true
+	case "r10":
+		return r.R10, true
+	case "r11":
+		return r.R11, true
+	case "r12":
+		return r.R12, true
+	case "r13":
+		return r.R13, true
+	case "r14":
+		return r.R14, true
+	case "r15":
+		return r.R15, true
+	case "rip", "eip":
+		return r.RIP, true
+	default:
+		return 0, false
+	}
+}
+
+// breakpointInsn is the software breakpoint instruction.
+var breakpointInsn = []byte{0xCC} // INT3
+
+// breakpointPCAdjust is how far PC is past the breakpoint instruction after
+// a trap.  INT3 leaves the hardware PC on the following instruction (Intel
+// SDM vol 3, "Debug Exceptions") and the FreeBSD T_BPTFLT path delivers
+// that frame unmodified (sys/amd64/amd64/trap.c), so the trap address is
+// PC-1, the same correction the Linux side applies.
+const breakpointPCAdjust = 1

debug/regs_freebsd_arm64.go

@@ -0,0 +1,139 @@
+// Copyright (c) 2026 Petr Balvín <opensource@petrbalvin.org> (https://petrbalvin.org)
+// SPDX-License-Identifier: BSD-3-Clause
+
+//go:build freebsd && arm64
+
+package debug
+
+// Regs holds the full general-purpose register set of a traced process
+// (the FreeBSD arm64 struct reg layout, sys/arm64/include/reg.h: x[30], lr,
+// sp, elr, spsr).
+type Regs struct {
+	X0     uint64
+	X1     uint64
+	X2     uint64
+	X3     uint64
+	X4     uint64
+	X5     uint64
+	X6     uint64
+	X7     uint64
+	X8     uint64
+	X9     uint64
+	X10    uint64
+	X11    uint64
+	X12    uint64
+	X13    uint64
+	X14    uint64
+	X15    uint64
+	X16    uint64
+	X17    uint64
+	X18    uint64
+	X19    uint64
+	X20    uint64
+	X21    uint64
+	X22    uint64
+	X23    uint64
+	X24    uint64
+	X25    uint64
+	X26    uint64
+	X27    uint64
+	X28    uint64
+	X29    uint64 // FP (frame pointer)
+	X30    uint64 // LR (link register)
+	SP     uint64
+	PC     uint64
+	PSTATE uint64
+}
+
+// GetPC returns the program counter.
+func (r *Regs) GetPC() uint64 { return r.PC }
+
+// SetPC sets the program counter.
+func (r *Regs) SetPC(pc uint64) { r.PC = pc }
+
+// GetSP returns the stack pointer.
+func (r *Regs) GetSP() uint64 { return r.SP }
+
+// RegValue returns the value of the named register, or false if unknown.
+func (r *Regs) RegValue(name string) (uint64, bool) {
+	switch name {
+	case "x0":
+		return r.X0, true
+	case "x1":
+		return r.X1, true
+	case "x2":
+		return r.X2, true
+	case "x3":
+		return r.X3, true
+	case "x4":
+		return r.X4, true
+	case "x5":
+		return r.X5, true
+	case "x6":
+		return r.X6, true
+	case "x7":
+		return r.X7, true
+	case "x8":
+		return r.X8, true
+	case "x9":
+		return r.X9, true
+	case "x10":
+		return r.X10, true
+	case "x11":
+		return r.X11, true
+	case "x12":
+		return r.X12, true
+	case "x13":
+		return r.X13, true
+	case "x14":
+		return r.X14, true
+	case "x15":
+		return r.X15, true
+	case "x16":
+		return r.X16, true
+	case "x17":
+		return r.X17, true
+	case "x18":
+		return r.X18, true
+	case "x19":
+		return r.X19, true
+	case "x20":
+		return r.X20, true
+	case "x21":
+		return r.X21, true
+	case "x22":
+		return r.X22, true
+	case "x23":
+		return r.X23, true
+	case "x24":
+		return r.X24, true
+	case "x25":
+		return r.X25, true
+	case "x26":
+		return r.X26, true
+	case "x27":
+		return r.X27, true
+	case "x28":
+		return r.X28, true
+	case "x29", "fp":
+		return r.X29, true
+	case "x30", "lr":
+		return r.X30, true
+	case "sp":
+		return r.SP, true
+	case "pc":
+		return r.PC, true
+	default:
+		return 0, false
+	}
+}
+
+// breakpointInsn is the software breakpoint instruction (BRK #0).
+var breakpointInsn = []byte{0x00, 0x00, 0x20, 0xD4} // BRK #0
+
+// breakpointPCAdjust is how far PC is past the breakpoint instruction after
+// a trap: 0.  The BRK synchronous exception leaves ELR_EL0 on the BRK
+// itself (ARM DDI 0487), and the FreeBSD EXCP_BRKPT_EL0 handler delivers
+// the frame's elr unmodified (sys/arm64/arm64/trap.c), so the trap address
+// is the PC as reported.
+const breakpointPCAdjust = 0

debug/regs_freebsd_riscv64.go

@@ -0,0 +1,134 @@
+// Copyright (c) 2026 Petr Balvín <opensource@petrbalvin.org> (https://petrbalvin.org)
+// SPDX-License-Identifier: BSD-3-Clause
+
+//go:build freebsd && riscv64
+
+package debug
+
+// Regs holds the full general-purpose register set of a traced process
+// (the FreeBSD riscv64 struct reg layout: ra, sp, gp, tp, t0-t6, s0-s11,
+// a0-a7, sepc, sstatus).
+type Regs struct {
+	PC  uint64 // sepc
+	Ra  uint64 // x1 (return address)
+	Sp  uint64 // x2
+	Gp  uint64 // x3
+	Tp  uint64 // x4
+	T0  uint64 // x5
+	T1  uint64 // x6
+	T2  uint64 // x7
+	S0  uint64 // x8 (frame pointer)
+	S1  uint64 // x9
+	A0  uint64 // x10
+	A1  uint64 // x11
+	A2  uint64 // x12
+	A3  uint64 // x13
+	A4  uint64 // x14
+	A5  uint64 // x15
+	A6  uint64 // x16
+	A7  uint64 // x17
+	S2  uint64 // x18
+	S3  uint64 // x19
+	S4  uint64 // x20
+	S5  uint64 // x21
+	S6  uint64 // x22
+	S7  uint64 // x23
+	S8  uint64 // x24
+	S9  uint64 // x25
+	S10 uint64 // x26
+	S11 uint64 // x27
+	T3  uint64 // x28
+	T4  uint64 // x29
+	T5  uint64 // x30
+	T6  uint64 // x31
+}
+
+// GetPC returns the program counter.
+func (r *Regs) GetPC() uint64 { return r.PC }
+
+// SetPC sets the program counter.
+func (r *Regs) SetPC(pc uint64) { r.PC = pc }
+
+// GetSP returns the stack pointer.
+func (r *Regs) GetSP() uint64 { return r.Sp }
+
+// RegValue returns the value of the named register, or false if unknown.
+func (r *Regs) RegValue(name string) (uint64, bool) {
+	switch name {
+	case "pc":
+		return r.PC, true
+	case "ra", "x1":
+		return r.Ra, true
+	case "sp", "x2":
+		return r.Sp, true
+	case "gp", "x3":
+		return r.Gp, true
+	case "tp", "x4":
+		return r.Tp, true
+	case "t0", "x5":
+		return r.T0, true
+	case "t1", "x6":
+		return r.T1, true
+	case "t2", "x7":
+		return r.T2, true
+	case "s0", "fp", "x8":
+		return r.S0, true
+	case "s1", "x9":
+		return r.S1, true
+	case "a0", "x10":
+		return r.A0, true
+	case "a1", "x11":
+		return r.A1, true
+	case "a2", "x12":
+		return r.A2, true
+	case "a3", "x13":
+		return r.A3, true
+	case "a4", "x14":
+		return r.A4, true
+	case "a5", "x15":
+		return r.A5, true
+	case "a6", "x16":
+		return r.A6, true
+	case "a7", "x17":
+		return r.A7, true
+	case "s2", "x18":
+		return r.S2, true
+	case "s3", "x19":
+		return r.S3, true
+	case "s4", "x20":
+		return r.S4, true
+	case "s5", "x21":
+		return r.S5, true
+	case "s6", "x22":
+		return r.S6, true
+	case "s7", "x23":
+		return r.S7, true
+	case "s8", "x24":
+		return r.S8, true
+	case "s9", "x25":
+		return r.S9, true
+	case "s10", "x26":
+		return r.S10, true
+	case "s11", "x27":
+		return r.S11, true
+	case "t3", "x28":
+		return r.T3, true
+	case "t4", "x29":
+		return r.T4, true
+	case "t5", "x30":
+		return r.T5, true
+	case "t6", "x31":
+		return r.T6, true
+	default:
+		return 0, false
+	}
+}
+
+// breakpointInsn is the software breakpoint instruction (EBREAK).
+var breakpointInsn = []byte{0x73, 0x00, 0x10, 0x00} // ebreak
+
+// breakpointPCAdjust is how far PC is past the breakpoint instruction after
+// a trap: 0.  The EBREAK synchronous exception leaves sepc on the ebreak
+// itself (RISC-V privileged architecture), so the trap address is the PC as
+// reported.
+const breakpointPCAdjust = 0

debug/repl.go

@@ -1,7 +1,7 @@
 // Copyright (c) 2026 Petr Balvín <opensource@petrbalvin.org> (https://petrbalvin.org)
 // SPDX-License-Identifier: BSD-3-Clause
 
-//go:build linux
+//go:build linux || (freebsd && (amd64 || arm64 || riscv64))
 
 package debug
 

debug/stopinfo_freebsd.go

@@ -0,0 +1,73 @@
+// Copyright (c) 2026 Petr Balvín <opensource@petrbalvin.org> (https://petrbalvin.org)
+// SPDX-License-Identifier: BSD-3-Clause
+
+//go:build freebsd && (amd64 || arm64 || riscv64)
+
+package debug
+
+import (
+	"encoding/binary"
+	"syscall"
+	"unsafe"
+
+	"golang.org/x/sys/unix"
+)
+
+// FreeBSD TRAP_* si_code values (sys/signal.h).  A breakpoint (INT3, BRK,
+// EBREAK) arrives as TRAP_BRKPT on every supported architecture; TRAP_TRACE
+// is shared by the completed single-step and the hardware watchpoint hit,
+// so the watchpoint layer disambiguates from the debug registers.
+const (
+	trapBRKPT = 1 // TRAP_BRKPT
+	trapTRACE = 2 // TRAP_TRACE
+)
+
+// StopReason describes why the debuggee stopped.
+type StopReason int
+
+const (
+	StopNone       StopReason = iota
+	StopBreakpoint            // software breakpoint hit
+	StopWatchpoint            // hardware watchpoint triggered
+	StopSingleStep            // single-step completed
+	StopSignal                // stopped by a signal
+	StopExited                // process exited
+)
+
+// StopInfo returns the reason the debuggee stopped and the faulting address
+// (for watchpoints, the watched address that was accessed).  FreeBSD has no
+// PTRACE_GETSIGINFO; the stop's signal information comes from PT_LWPINFO,
+// whose pl_siginfo carries the siginfo the kernel delivered.  A ptrace stop
+// with no signal behind it (a completed single-step, the initial attach)
+// fills no siginfo at all.
+func (s *Session) StopInfo() (StopReason, uint64) {
+	if s.exited {
+		return StopExited, 0
+	}
+	var info unix.PtraceLwpInfoStruct
+	if err := unix.PtraceLwpInfo(s.pid, &info); err != nil {
+		return StopNone, 0
+	}
+	// The siginfo layout is the FreeBSD siginfo_t: three leading ints
+	// (signo, errno, code), then the union, 8-byte aligned, whose _fault
+	// member puts the address at byte offset 16.  The read is byte-wise
+	// because the blob's alignment is not guaranteed.
+	si := (*[64]byte)(unsafe.Pointer(&info.Siginfo))
+	signo := int32(binary.LittleEndian.Uint32(si[0:4]))
+	code := int32(binary.LittleEndian.Uint32(si[8:12]))
+	switch {
+	case signo == 0:
+		// A pure ptrace stop: single-step completion, attach, or the
+		// events the kernel resolves internally.
+		return StopSingleStep, 0
+	case signo != int32(syscall.SIGTRAP):
+		return StopSignal, uint64(code)
+	case code == trapBRKPT:
+		return StopBreakpoint, 0
+	case code == trapTRACE:
+		addr := binary.LittleEndian.Uint64(si[16:24])
+		return archStopTrace(s, addr)
+	default:
+		return StopSingleStep, 0
+	}
+}

debug/target_freebsd.go

@@ -0,0 +1,195 @@
+// Copyright (c) 2026 Petr Balvín <opensource@petrbalvin.org> (https://petrbalvin.org)
+// SPDX-License-Identifier: BSD-3-Clause
+
+//go:build freebsd && (amd64 || arm64 || riscv64)
+
+package debug
+
+import (
+	"encoding/hex"
+	"fmt"
+	"os"
+	"runtime"
+	"strconv"
+	"strings"
+	"syscall"
+	"unsafe"
+
+	"golang.org/x/sys/unix"
+
+	"sourcedock.dev/petrbalvin/gasm-sdk/asm"
+	"sourcedock.dev/petrbalvin/gasm-sdk/parser"
+	"sourcedock.dev/petrbalvin/gasm-sdk/verify"
+)
+
+// mapRWX maps code into a read-write-execute region.
+func mapRWX(code []byte) ([]byte, error) {
+	const pageSize = 4096
+	size := (len(code) + pageSize - 1) &^ (pageSize - 1)
+	mem, err := syscall.Mmap(-1, 0, size,
+		syscall.PROT_READ|syscall.PROT_WRITE|syscall.PROT_EXEC,
+		syscall.MAP_PRIVATE|syscall.MAP_ANON)
+	if err != nil {
+		return nil, err
+	}
+	copy(mem, code)
+	return mem, nil
+}
+
+// setupBuffers allocates buffers in the debuggee's memory.
+func setupBuffers(spec string, args []byte, tmpDir string) ([]byte, error) {
+	type bufSpec struct {
+		name    string
+		size    int
+		pattern string
+	}
+	var specs []bufSpec
+	for part := range strings.SplitSeq(spec, ",") {
+		fields := strings.SplitN(part, ":", 3)
+		if len(fields) != 3 {
+			continue
+		}
+		size, err := strconv.Atoi(fields[1])
+		if err != nil || size <= 0 {
+			continue
+		}
+		specs = append(specs, bufSpec{name: fields[0], size: size, pattern: fields[2]})
+	}
+
+	if len(specs) == 0 {
+		return args, nil
+	}
+
+	var bufAddrs []uint64
+	for _, s := range specs {
+		buf, err := syscall.Mmap(-1, 0, s.size,
+			syscall.PROT_READ|syscall.PROT_WRITE,
+			syscall.MAP_PRIVATE|syscall.MAP_ANON)
+		if err != nil {
+			return nil, fmt.Errorf("mmap buffer %s: %w", s.name, err)
+		}
+		fillBuffer(buf, s.pattern)
+		bufAddrs = append(bufAddrs, uint64(uintptr(unsafe.Pointer(&buf[0]))))
+	}
+
+	addrFile, err := os.Create(tmpDir + "/bufaddrs")
+	if err != nil {
+		return nil, err
+	}
+	for _, addr := range bufAddrs {
+		fmt.Fprintf(addrFile, "%d\n", addr)
+	}
+	addrFile.Close()
+
+	return args, nil
+}
+
+// fillBuffer fills a buffer with the specified pattern.
+func fillBuffer(buf []byte, pattern string) {
+	switch pattern {
+	case "zero":
+	case "ones":
+		for i := range buf {
+			buf[i] = 0xFF
+		}
+	case "seq":
+		for i := range buf {
+			buf[i] = byte(i)
+		}
+	default:
+		if data, err := hex.DecodeString(pattern); err == nil && len(data) > 0 {
+			for i := range buf {
+				buf[i] = data[i%len(data)]
+			}
+		}
+	}
+}
+
+// RunTarget is the debuggee entry point (gasm debug --target).
+func RunTarget(asmPath, funcName, argsFile, tmpDir string) error {
+	src, err := os.ReadFile(asmPath)
+	if err != nil {
+		return fmt.Errorf("debug target: %w", err)
+	}
+	file, errs := parser.Parse(asmPath, string(src))
+	if len(errs) > 0 {
+		return fmt.Errorf("debug target: parse: %v", errs[0])
+	}
+	img, err := asm.AssembleFile(file)
+	if err != nil {
+		return fmt.Errorf("debug target: assemble: %w", err)
+	}
+
+	var fl *asm.FuncLayout
+	for i := range img.Funcs {
+		if img.Funcs[i].Name == funcName {
+			fl = &img.Funcs[i]
+			break
+		}
+	}
+	if fl == nil {
+		return fmt.Errorf("debug target: function %q not found", funcName)
+	}
+
+	code := img.Bytes()
+	exec, err := mapRWX(code)
+	if err != nil {
+		return fmt.Errorf("debug target: mmap: %w", err)
+	}
+
+	codeBase := uintptr(unsafe.Pointer(&exec[0]))
+	if err := os.WriteFile(tmpDir+"/codebase", []byte(fmt.Sprintf("%d", codeBase)), 0o644); err != nil {
+		return fmt.Errorf("debug target: write codebase: %w", err)
+	}
+
+	meta := fmt.Sprintf("%d %d %d", fl.Offset, fl.Size, fl.Args)
+	os.WriteFile(tmpDir+"/funcmeta", []byte(meta), 0o644)
+
+	labelsFile, _ := os.Create(tmpDir + "/labels")
+	if labelsFile != nil {
+		for label, off := range fl.Labels {
+			fmt.Fprintf(labelsFile, "%s %d\n", label, off)
+		}
+		labelsFile.Close()
+	}
+
+	args, err := os.ReadFile(argsFile)
+	if err != nil {
+		return fmt.Errorf("debug target: read args: %w", err)
+	}
+	if len(args) < fl.Args {
+		padded := make([]byte, fl.Args)
+		copy(padded, args)
+		args = padded
+	}
+
+	bufSpecFile := tmpDir + "/bufspec"
+	if bufSpec, err := os.ReadFile(bufSpecFile); err == nil && len(bufSpec) > 0 {
+		args, err = setupBuffers(string(bufSpec), args, tmpDir)
+		if err != nil {
+			return fmt.Errorf("debug target: setup buffers: %w", err)
+		}
+	}
+
+	runtime.LockOSThread()
+
+	if _, _, errno := unix.RawSyscall(unix.SYS_PTRACE, uintptr(unix.PT_TRACE_ME), 0, 0); errno != 0 {
+		return fmt.Errorf("debug target: PT_TRACE_ME: %v", errno)
+	}
+	os.WriteFile(tmpDir+"/ready", []byte("ok"), 0o644)
+	syscall.Kill(syscall.Getpid(), syscall.SIGSTOP)
+
+	os.WriteFile(tmpDir+"/entry", []byte("ok"), 0o644)
+	syscall.Kill(syscall.Getpid(), syscall.SIGSTOP)
+
+	fnAddr := codeBase + uintptr(fl.Offset)
+	stackArgs := make([]byte, fl.Args)
+	copy(stackArgs, args)
+
+	if _, callErr := verify.Call(fnAddr, stackArgs); callErr != nil {
+		os.Exit(1)
+	}
+	// Success returns to the caller, which exits with status 0; the JIT
+	// code has already run to its own trampoline by the time Call returns.
+	return nil
+}

debug/target_linux_amd64.go

@@ -12,9 +12,9 @@ import (
 	"syscall"
 	"unsafe"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/asm"
-	"sourcedock.dev/petrbalvin/gasm-devkit/parser"
-	"sourcedock.dev/petrbalvin/gasm-devkit/verify"
+	"sourcedock.dev/petrbalvin/gasm-sdk/asm"
+	"sourcedock.dev/petrbalvin/gasm-sdk/parser"
+	"sourcedock.dev/petrbalvin/gasm-sdk/verify"
 )
 
 // RunTarget is the debuggee entry point (gasm debug --target).

debug/target_linux_arm64.go

@@ -12,9 +12,9 @@ import (
 	"syscall"
 	"unsafe"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/asm"
-	"sourcedock.dev/petrbalvin/gasm-devkit/parser"
-	"sourcedock.dev/petrbalvin/gasm-devkit/verify"
+	"sourcedock.dev/petrbalvin/gasm-sdk/asm"
+	"sourcedock.dev/petrbalvin/gasm-sdk/parser"
+	"sourcedock.dev/petrbalvin/gasm-sdk/verify"
 )
 
 // RunTarget is the debuggee entry point (gasm debug --target).

debug/target_linux_loong64.go

@@ -12,9 +12,9 @@ import (
 	"syscall"
 	"unsafe"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/asm"
-	"sourcedock.dev/petrbalvin/gasm-devkit/parser"
-	"sourcedock.dev/petrbalvin/gasm-devkit/verify"
+	"sourcedock.dev/petrbalvin/gasm-sdk/asm"
+	"sourcedock.dev/petrbalvin/gasm-sdk/parser"
+	"sourcedock.dev/petrbalvin/gasm-sdk/verify"
 )
 
 // RunTarget is the debuggee entry point (gasm debug --target).

debug/target_linux_riscv64.go

@@ -12,9 +12,9 @@ import (
 	"syscall"
 	"unsafe"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/asm"
-	"sourcedock.dev/petrbalvin/gasm-devkit/parser"
-	"sourcedock.dev/petrbalvin/gasm-devkit/verify"
+	"sourcedock.dev/petrbalvin/gasm-sdk/asm"
+	"sourcedock.dev/petrbalvin/gasm-sdk/parser"
+	"sourcedock.dev/petrbalvin/gasm-sdk/verify"
 )
 
 // RunTarget is the debuggee entry point (gasm debug --target).

debug/tracer.go

@@ -1,7 +1,7 @@
 // Copyright (c) 2026 Petr Balvín <opensource@petrbalvin.org> (https://petrbalvin.org)
 // SPDX-License-Identifier: BSD-3-Clause
 
-//go:build linux
+//go:build linux || (freebsd && (amd64 || arm64 || riscv64))
 
 package debug
 

debug/watchpoint_freebsd_amd64.go

@@ -0,0 +1,190 @@
+// Copyright (c) 2026 Petr Balvín <opensource@petrbalvin.org> (https://petrbalvin.org)
+// SPDX-License-Identifier: BSD-3-Clause
+
+//go:build freebsd && amd64
+
+package debug
+
+import (
+	"fmt"
+	"unsafe"
+
+	"golang.org/x/sys/unix"
+)
+
+// Hardware watchpoint support via x86-64 debug registers (DR0-DR3, DR7),
+// read and written as one blob through PT_GETDBREGS/PT_SETDBREGS.  The
+// FreeBSD struct dbreg is the raw DR file: dr[16], where DR0-DR3 are the
+// address registers, DR6 the status and DR7 the control (sys/x86/include/
+// reg.h; the DBREG_DRX accessor indexes the same array).
+
+// dbreg mirrors FreeBSD's struct dbreg for PT_GETDBREGS/PT_SETDBREGS.
+type dbreg struct {
+	Dr [16]uint64
+}
+
+// dbreg indices of the registers the watchpoint layer drives.
+const (
+	drStatus  = 6 // DR6: the trap status register
+	drControl = 7 // DR7: the debug control register
+)
+
+// WatchpointType selects what triggers the watchpoint.
+type WatchpointType int
+
+const (
+	WatchWrite WatchpointType = 1 // trigger on write
+	WatchRead  WatchpointType = 3 // trigger on read or write
+)
+
+// maxWatchpoints reports the number of hardware watchpoint slots the
+// architecture provides: four address registers, DR0-DR3.
+func maxWatchpoints() int { return 4 }
+
+// getDbRegs reads the debug register file of the stopped debuggee.
+func (s *Session) getDbRegs() (*dbreg, error) {
+	var dr dbreg
+	if _, _, errno := unix.Syscall6(
+		unix.SYS_PTRACE,
+		uintptr(unix.PT_GETDBREGS),
+		uintptr(s.pid),
+		0,
+		uintptr(unsafe.Pointer(&dr)),
+		0, 0,
+	); errno != 0 {
+		return nil, errno
+	}
+	return &dr, nil
+}
+
+// setDbRegs writes the debug register file of the stopped debuggee.
+func (s *Session) setDbRegs(dr *dbreg) error {
+	if _, _, errno := unix.Syscall6(
+		unix.SYS_PTRACE,
+		uintptr(unix.PT_SETDBREGS),
+		uintptr(s.pid),
+		0,
+		uintptr(unsafe.Pointer(dr)),
+		0, 0,
+	); errno != 0 {
+		return errno
+	}
+	return nil
+}
+
+// archStopTrace classifies a TRAP_TRACE stop.  On amd64 the kernel
+// delivers both the completed single-step and the debug-register hit
+// through T_TRCTRAP with TRAP_TRACE (sys/amd64/amd64/trap.c), and DR6's
+// B0-B3 bits name the watchpoint that fired.
+func archStopTrace(s *Session, siAddr uint64) (StopReason, uint64) {
+	dr, err := s.getDbRegs()
+	if err != nil {
+		return StopSingleStep, 0
+	}
+	if status := dr.Dr[drStatus]; status&0xF != 0 {
+		for slot := range 4 {
+			if status&(1<<slot) != 0 && dr.Dr[slot] != 0 {
+				return StopWatchpoint, dr.Dr[slot]
+			}
+		}
+	}
+	return StopSingleStep, 0
+}
+
+// FindFreeWatchpointSlot returns the index of the first free watchpoint slot
+// (0-3), or -1 if all four hardware watchpoints are in use.
+func (s *Session) FindFreeWatchpointSlot() int {
+	for i := range 4 {
+		if !s.wpSlots[i] {
+			return i
+		}
+	}
+	return -1
+}
+
+// IsWatchpointSlotUsed reports whether slot (0-3) currently holds a watchpoint.
+func (s *Session) IsWatchpointSlotUsed(slot int) bool {
+	if slot < 0 || slot > 3 {
+		return false
+	}
+	return s.wpSlots[slot]
+}
+
+// SetWatchpoint installs a hardware watchpoint on the given address.
+// DR7's encoding is architectural: a 2-bit local/global enable pair per
+// slot at bit 2*slot, the R/W field at 16+4*slot and the length field at
+// 18+4*slot (Intel SDM vol 3, "Debug Registers").
+func (s *Session) SetWatchpoint(slot int, addr uint64, typ WatchpointType, size int) error {
+	if slot < 0 || slot > 3 {
+		return fmt.Errorf("debug: watchpoint slot must be 0-3")
+	}
+	if s.wpSlots[slot] {
+		return fmt.Errorf("debug: watchpoint slot %d already in use", slot)
+	}
+
+	var lenBits uint64
+	switch size {
+	case 1:
+		lenBits = 0
+	case 2:
+		lenBits = 1
+	case 4:
+		lenBits = 3
+	case 8:
+		lenBits = 2
+	default:
+		return fmt.Errorf("debug: watchpoint size must be 1, 2, 4, or 8")
+	}
+
+	dr, err := s.getDbRegs()
+	if err != nil {
+		return fmt.Errorf("debug: read debug registers: %w", err)
+	}
+	dr.Dr[slot] = addr
+
+	dr7 := dr.Dr[drControl]
+	enableBit := uint64(1) << (2 * slot)
+	rwBits := uint64(typ) << (16 + 4*slot)
+	lenField := lenBits << (18 + 4*slot)
+	mask := ^((uint64(1) << (2 * slot)) | (uint64(3) << (16 + 4*slot)) | (uint64(3) << (18 + 4*slot)))
+	dr.Dr[drControl] = (dr7 & mask) | enableBit | rwBits | lenField
+
+	if err := s.setDbRegs(dr); err != nil {
+		return fmt.Errorf("debug: set debug registers: %w", err)
+	}
+	s.wpSlots[slot] = true
+	return nil
+}
+
+// ClearWatchpoint removes a hardware watchpoint.
+func (s *Session) ClearWatchpoint(slot int) error {
+	if slot < 0 || slot > 3 {
+		return fmt.Errorf("debug: watchpoint slot must be 0-3")
+	}
+	if !s.wpSlots[slot] {
+		return fmt.Errorf("debug: watchpoint slot %d is not in use", slot)
+	}
+	dr, err := s.getDbRegs()
+	if err != nil {
+		return err
+	}
+	dr.Dr[slot] = 0
+	dr.Dr[drControl] &^= uint64(1) << (2 * slot)
+	if err := s.setDbRegs(dr); err != nil {
+		return err
+	}
+	s.wpSlots[slot] = false
+	return nil
+}
+
+// ClearAllWatchpoints removes all hardware watchpoints.
+func (s *Session) ClearAllWatchpoints() error {
+	for slot := range maxWatchpoints() {
+		if s.wpSlots[slot] {
+			if err := s.ClearWatchpoint(slot); err != nil {
+				return err
+			}
+		}
+	}
+	return nil
+}

debug/watchpoint_freebsd_arm64.go

@@ -0,0 +1,205 @@
+// Copyright (c) 2026 Petr Balvín <opensource@petrbalvin.org> (https://petrbalvin.org)
+// SPDX-License-Identifier: BSD-3-Clause
+
+//go:build freebsd && arm64
+
+package debug
+
+import (
+	"fmt"
+	"unsafe"
+
+	"golang.org/x/sys/unix"
+)
+
+// Hardware watchpoint support via arm64 debug registers, read and written
+// as one blob through PT_GETDBREGS/PT_SETDBREGS.  The FreeBSD struct dbreg
+// (sys/arm64/include/reg.h) opens with the debug-facility header and then
+// carries 16 breakpoint and 16 watchpoint pairs of {address, control}.
+
+// dbreg mirrors FreeBSD's struct dbreg for PT_GETDBREGS/PT_SETDBREGS.
+type dbreg struct {
+	DbDebugVer  uint8
+	DbNbkpts    uint8
+	DbNwtpts    uint8
+	_           [5]byte
+	DbBreakregs [16]struct {
+		Addr uint64
+		Ctrl uint32
+		_    uint32
+	}
+	DbWatchregs [16]struct {
+		Addr uint64
+		Ctrl uint32
+		_    uint32
+	}
+}
+
+// WatchpointType selects what triggers the watchpoint.
+type WatchpointType int
+
+const (
+	WatchWrite WatchpointType = 1 // trigger on write
+	WatchRead  WatchpointType = 3 // trigger on read or write
+)
+
+// maxWatchpoints reports the number of hardware watchpoint slots the
+// architecture provides: DBGWVR0-DBGWCR15.
+func maxWatchpoints() int { return 16 }
+
+// getDbRegs reads the debug register file of the stopped debuggee.
+func (s *Session) getDbRegs() (*dbreg, error) {
+	var dr dbreg
+	if _, _, errno := unix.Syscall6(
+		unix.SYS_PTRACE,
+		uintptr(unix.PT_GETDBREGS),
+		uintptr(s.pid),
+		0,
+		uintptr(unsafe.Pointer(&dr)),
+		0, 0,
+	); errno != 0 {
+		return nil, errno
+	}
+	return &dr, nil
+}
+
+// setDbRegs writes the debug register file of the stopped debuggee.
+func (s *Session) setDbRegs(dr *dbreg) error {
+	if _, _, errno := unix.Syscall6(
+		unix.SYS_PTRACE,
+		uintptr(unix.PT_SETDBREGS),
+		uintptr(s.pid),
+		0,
+		uintptr(unsafe.Pointer(dr)),
+		0, 0,
+	); errno != 0 {
+		return errno
+	}
+	return nil
+}
+
+// archStopTrace classifies a TRAP_TRACE stop.  On arm64 the kernel
+// delivers both the software single step and the watchpoint hit through
+// EXCP_SOFTSTP_EL0/EXCP_WATCHPT_EL0 with TRAP_TRACE (sys/arm64/arm64/
+// trap.c); the watchpoint address rides the FAR register, so a stop whose
+// reported address falls inside an armed watchpoint's byte range is a
+// watchpoint and everything else is a single step.
+func archStopTrace(s *Session, siAddr uint64) (StopReason, uint64) {
+	dr, err := s.getDbRegs()
+	if err != nil {
+		return StopSingleStep, 0
+	}
+	for slot := range 16 {
+		ctrl := uint64(dr.DbWatchregs[slot].Ctrl)
+		if ctrl&1 == 0 || dr.DbWatchregs[slot].Addr == 0 {
+			continue
+		}
+		if bas := (ctrl >> 5) & 0xFF; bas != 0 && siAddr >= dr.DbWatchregs[slot].Addr && siAddr < dr.DbWatchregs[slot].Addr+8 {
+			return StopWatchpoint, siAddr
+		}
+	}
+	return StopSingleStep, 0
+}
+
+// FindFreeWatchpointSlot returns the index of the first free watchpoint
+// slot, or -1 if all of them are in use.
+func (s *Session) FindFreeWatchpointSlot() int {
+	for i := range maxWatchpoints() {
+		if !s.wpSlots[i] {
+			return i
+		}
+	}
+	return -1
+}
+
+// IsWatchpointSlotUsed reports whether slot currently holds a watchpoint.
+func (s *Session) IsWatchpointSlotUsed(slot int) bool {
+	if slot < 0 || slot >= maxWatchpoints() {
+		return false
+	}
+	return s.wpSlots[slot]
+}
+
+// SetWatchpoint installs a hardware watchpoint on the given address.  The
+// control word is the architectural DBGWCR (ARM DDI 0487): bit 0 enables,
+// bits 3-4 select the access type (10 store, 11 load+store) and bits 5-12
+// are the byte-address select, so the watch stays 8-byte aligned and names
+// its watched bytes through BAS.
+func (s *Session) SetWatchpoint(slot int, addr uint64, typ WatchpointType, size int) error {
+	if slot < 0 || slot >= maxWatchpoints() {
+		return fmt.Errorf("debug: watchpoint slot must be 0-%d", maxWatchpoints()-1)
+	}
+	if s.wpSlots[slot] {
+		return fmt.Errorf("debug: watchpoint slot %d already in use", slot)
+	}
+	var bas uint64
+	switch size {
+	case 1:
+		bas = 0x01
+	case 2:
+		bas = 0x03
+	case 4:
+		bas = 0x0F
+	case 8:
+		bas = 0xFF
+	default:
+		return fmt.Errorf("debug: watchpoint size must be 1, 2, 4, or 8")
+	}
+
+	dr, err := s.getDbRegs()
+	if err != nil {
+		return fmt.Errorf("debug: read debug registers: %w", err)
+	}
+	if uint8(slot) >= dr.DbNwtpts && dr.DbNwtpts != 0 {
+		return fmt.Errorf("debug: slot %d exceeds available watchpoints (%d)", slot, dr.DbNwtpts)
+	}
+	ctrl := uint64(1) // enable
+	switch typ {
+	case WatchWrite:
+		ctrl |= 2 << 3 // store only
+	case WatchRead:
+		ctrl |= 3 << 3 // load+store
+	}
+	ctrl |= bas << 5
+	dr.DbWatchregs[slot].Addr = addr
+	dr.DbWatchregs[slot].Ctrl = uint32(ctrl)
+
+	if err := s.setDbRegs(dr); err != nil {
+		return fmt.Errorf("debug: set debug registers: %w", err)
+	}
+	s.wpSlots[slot] = true
+	return nil
+}
+
+// ClearWatchpoint removes a hardware watchpoint.
+func (s *Session) ClearWatchpoint(slot int) error {
+	if slot < 0 || slot >= maxWatchpoints() {
+		return fmt.Errorf("debug: watchpoint slot must be 0-%d", maxWatchpoints()-1)
+	}
+	if !s.wpSlots[slot] {
+		return fmt.Errorf("debug: watchpoint slot %d is not in use", slot)
+	}
+	dr, err := s.getDbRegs()
+	if err != nil {
+		return err
+	}
+	dr.DbWatchregs[slot].Addr = 0
+	dr.DbWatchregs[slot].Ctrl = 0
+	if err := s.setDbRegs(dr); err != nil {
+		return err
+	}
+	s.wpSlots[slot] = false
+	return nil
+}
+
+// ClearAllWatchpoints removes all hardware watchpoints.
+func (s *Session) ClearAllWatchpoints() error {
+	for slot := range maxWatchpoints() {
+		if s.wpSlots[slot] {
+			if err := s.ClearWatchpoint(slot); err != nil {
+				return err
+			}
+		}
+	}
+	return nil
+}

debug/watchpoint_freebsd_riscv64.go

@@ -0,0 +1,50 @@
+// Copyright (c) 2026 Petr Balvín <opensource@petrbalvin.org> (https://petrbalvin.org)
+// SPDX-License-Identifier: BSD-3-Clause
+
+//go:build freebsd && riscv64
+
+package debug
+
+import "fmt"
+
+// The architecture has hardware watchpoint triggers, but FreeBSD exposes
+// no PT_GETDBREGS request for riscv64, so there is no supported way to arm
+// one: the watchpoint layer is honestly empty here.
+
+// WatchpointType selects what triggers the watchpoint.
+type WatchpointType int
+
+const (
+	WatchWrite WatchpointType = 1 // trigger on write
+	WatchRead  WatchpointType = 3 // trigger on read or write
+)
+
+// maxWatchpoints reports the number of hardware watchpoint slots the
+// platform provides: FreeBSD exposes none for riscv64.
+func maxWatchpoints() int { return 0 }
+
+// archStopTrace classifies a TRAP_TRACE stop; with no watchpoint layer a
+// trace stop is always a completed single step.
+func archStopTrace(s *Session, siAddr uint64) (StopReason, uint64) {
+	return StopSingleStep, 0
+}
+
+// FindFreeWatchpointSlot returns -1: no slots exist.
+func (s *Session) FindFreeWatchpointSlot() int { return -1 }
+
+// IsWatchpointSlotUsed reports whether slot currently holds a watchpoint.
+func (s *Session) IsWatchpointSlotUsed(slot int) bool { return false }
+
+// SetWatchpoint is unsupported: FreeBSD exposes no debug register request
+// for riscv64.
+func (s *Session) SetWatchpoint(slot int, addr uint64, typ WatchpointType, size int) error {
+	return fmt.Errorf("debug: hardware watchpoints are not supported on freebsd/riscv64")
+}
+
+// ClearWatchpoint is unsupported for the same reason.
+func (s *Session) ClearWatchpoint(slot int) error {
+	return fmt.Errorf("debug: hardware watchpoints are not supported on freebsd/riscv64")
+}
+
+// ClearAllWatchpoints is a no-op: no watchpoint can be armed.
+func (s *Session) ClearAllWatchpoints() error { return nil }

disasm/disasm.go

@@ -15,7 +15,7 @@ import (
 	"golang.org/x/arch/riscv64/riscv64asm"
 	"golang.org/x/arch/x86/x86asm"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/arch"
+	"sourcedock.dev/petrbalvin/gasm-sdk/arch"
 )
 
 // Instruction is one decoded instruction: its text form, its length in bytes

disasm/disasm_test.go

@@ -7,10 +7,10 @@ import (
 	"strings"
 	"testing"
 
-	"sourcedock.dev/petrbalvin/gasm-devkit/arch"
-	"sourcedock.dev/petrbalvin/gasm-devkit/asm"
-	"sourcedock.dev/petrbalvin/gasm-devkit/ast"
-	"sourcedock.dev/petrbalvin/gasm-devkit/parser"
+	"sourcedock.dev/petrbalvin/gasm-sdk/arch"
+	"sourcedock.dev/petrbalvin/gasm-sdk/asm"
+	"sourcedock.dev/petrbalvin/gasm-sdk/ast"
+	"sourcedock.dev/petrbalvin/gasm-sdk/parser"
 )
 
 func TestDecodeKnownBytes(t *testing.T) {

docs/ARCHITECTURE.md

@@ -1,8 +1,8 @@
 # Architecture
 
-How gasm-devkit is put together and why.
+How gasm-sdk is put together and why.
 
-Repository: [sourcedock.dev/petrbalvin/gasm-devkit](https://sourcedock.dev/petrbalvin/gasm-devkit)
+Repository: [sourcedock.dev/petrbalvin/gasm-sdk](https://sourcedock.dev/petrbalvin/gasm-sdk)
 
 ## Overview
 
@@ -119,6 +119,20 @@ identifier is a register or a label is an *architecture* question, so it is
 left to `arch` and resolved in the lint/lsp layers.  This keeps the parser
 arch-agnostic and its output deterministic.
 
+### Optional preprocessing
+
+With `Options{Expand: true}` the parser runs a pre-parse pass
+(`preproc.go`) that splices `#include` files (the source directory, then the
+`-I` directories), expands object and parameterised `#define` macros,
+applies `#undef` and the `#ifdef`/`#ifndef`/`#else`/`#endif` family, and
+folds constant expressions left in operands.  The go command's platform
+macros (`GOARCH_<arch>`, `GOOS_<goos>`) arrive through `Options.Predefines`.
+The assembly path (`asm`, `diff`, `audit`) expands; `lint`, `fmt` and the
+language server read the raw file.  The command layer adds the go_asm.h
+generator (`asmhdr.go`): a file that includes go_asm.h gets the package's
+defines type-checked out of its Go files for the target architecture and
+GOOS, with no compiler in the loop.
+
 ### `arch`
 
 Register files are generated programmatically (the regular `R8`-`R15`,
@@ -232,11 +246,18 @@ document store, republishes diagnostics on every change, and provides:
   pseudo-registers, labels, immediates, comments, directives, textflag macros);
 - **navigation**: go-to-definition from a label reference to its definition,
   find references, document highlights of every use of the symbol under the
-  cursor, rename, and workspace symbol search over the open documents;
+  cursor, rename, and workspace symbol search over the open documents and
+  the indexed workspace files: the `.s` files under the workspace root that
+  the editor has never opened, where an open buffer shadows its disk copy
+  and watched-file events plus a per-query freshness check keep the index
+  current;
 - **assists**: document formatting through the `format` package, inlay hints
   (the frame size after the TEXT argument area), signature help (the callee's
   `// func` signature while the cursor is on a `CALL`), and code actions
-  offering quick fixes for the `missing-ret` and `unused-label` diagnostics.
+  offering quick fixes for the `missing-ret` and `unused-label` diagnostics,
+  the `missing-textflag-include` warning (the include after the last one in
+  the file) and the `abi-argsize` warning (the argument area set to the size
+  the `// func` signature implies).
 - **document information**: pull diagnostics (`textDocument/diagnostic`),
   #include document links (resolved against the document directory, then
   `$GOROOT/pkg/include`) and folding ranges (one collapsible region per
@@ -558,7 +579,9 @@ AST, so neither depends on an encoding.
   --ground-truth`, `go list -json -export` locates the archives of the packages
   a GOOBJ object references, and `_gen` parses
   `$GOROOT/src/cmd/internal/obj/<arch>/anames.go` to rebuild the tables.
-- **Linux process interfaces** for the dynamic work: `mmap` and `mprotect` for
-  the JIT mapping, ptrace with `/proc/pid/mem` for the debugger.  That is why
+- **Linux and FreeBSD process interfaces** for the dynamic work: `mmap` and
+  `mprotect` for the JIT mapping, ptrace for the debugger — with tracee
+  memory through `/proc/pid/mem` on Linux and through `PT_IO` on FreeBSD,
+  and the tracee's stop reports read from `PT_LWPINFO` there.  That is why
   `verify` runs a JIT check only when the host architecture matches the
-  kernel's, and why `debug` is Linux-only.
+  kernel's, and why `debug` is bounded to those two kernels.

docs/CLI.md

@@ -287,7 +287,8 @@ Usage: gasm debug <file.s> --func <name>
 The debugger re-executes the binary it is running as (`os.Executable()`) for the
 traced child, so the child is the same `gasm`, whether it is installed on `$PATH`
 or run with `go run ./cmd/gasm`; nothing has to be installed first. Requires
-Linux (ptrace), and all four architectures are supported.
+Linux or FreeBSD (ptrace): all four architectures on Linux, amd64, arm64 and
+riscv64 on FreeBSD.
 
 REPL commands:
 
@@ -365,7 +366,7 @@ add: 16 bytes, args=24, frame=0 NOSPLIT
 ## audit-instructions
 
 ```text
-Usage: gasm audit-instructions [--corpus [dir]] [-I dir] [amd64|arm64|riscv64|loong64]
+Usage: gasm audit-instructions [--corpus [dir]] [--list] [-I dir] [amd64|arm64|riscv64|loong64]
 ```
 
 Compare the gasm encoder for the given architecture (default amd64) against the
@@ -403,7 +404,9 @@ architecture; a file without one is attempted for all four, exactly as a
 The report gives the headline number (files
 that assemble for every target architecture), the per-architecture pass rates
 and the most common failure reasons with one representative file each, which
-drive the encodability backlog by frequency rather than by table order. A run
+drive the encodability backlog by frequency rather than by table order. With
+`--list` the report additionally prints every failing file with its failure
+reason, per architecture. A run
 over GOROOT takes under a second.
 
 ```sh
@@ -453,7 +456,13 @@ semantic tokens, go-to-definition, find references, rename, document
 formatting, inlay hints, code actions, signature help, document highlights,
 workspace symbol search, #include document links, and folding ranges for
 function bodies. Definition, references and rename work across every open
-document.
+document and the wider workspace on disk: the server indexes the `.s` files
+under the workspace root that the editor has never opened, an open buffer
+always shadows its disk copy, and watched-file events together with a
+per-query freshness check keep the index current. The quick fixes add the
+missing `#include "textflag.h"`, set the TEXT argument area to the size the
+`// func` signature implies, add a missing `RET`, and remove an unused
+label.
 
 ## version
 

docs/DEVELOPMENT.md

@@ -1,6 +1,6 @@
 # Development Guide
 
-Repository: [sourcedock.dev/petrbalvin/gasm-devkit](https://sourcedock.dev/petrbalvin/gasm-devkit)
+Repository: [sourcedock.dev/petrbalvin/gasm-sdk](https://sourcedock.dev/petrbalvin/gasm-sdk)
 
 ## Prerequisites
 
@@ -19,8 +19,8 @@ Repository: [sourcedock.dev/petrbalvin/gasm-devkit](https://sourcedock.dev/petrb
 ## Setup
 
 ```sh
-git clone https://sourcedock.dev/petrbalvin/gasm-devkit.git
-cd gasm-devkit
+git clone https://sourcedock.dev/petrbalvin/gasm-sdk.git
+cd gasm-sdk
 just build     # compile bin/gasm, zero errors and zero warnings
 just gates     # build, fmt-check, vet, test, race: the definition of done
 ```

docs/GOOBJ.md

@@ -11,7 +11,7 @@ internal interface with no stability promise, and it can change in any
 release. This specification was therefore produced by reverse engineering
 that source and by parsing real objects produced by `go tool asm` and
 `go tool compile`, byte for byte, against the layout described here. Within
-gasm-devkit it is kept honest by the differential tests in `asm/goobj_test.go`
+gasm-sdk it is kept honest by the differential tests in `asm/goobj_test.go`
 and `asm/link_test.go`, which compare `gasm asm --format goobj` output against
 the toolchain's own products and feed gasm objects to `go build`.
 
@@ -615,7 +615,7 @@ objects, in one place:
 8. Compute the content hashes exactly as the toolchain does, or emit no
    hashed definitions at all.
 
-## How gasm-devkit implements and verifies it
+## How gasm-sdk implements and verifies it
 
 The writer lives in `asm/goobj.go`, which carries the shared container and the
 amd64 relocation emission, with per-architecture relocation emitters in

docs/asm/AMD64.md

@@ -0,0 +1,124 @@
+# AMD64
+
+Layer 1, target page. Verified against `go tool asm` of Go 1.27.1 and against
+gasm's encoder, whose output is compared byte for byte with the toolchain's
+and executed on real hardware (`gasm verify`). The complete mnemonic
+inventory lives in the generated appendix
+[INSTRUCTIONS-AMD64.md](INSTRUCTIONS-AMD64.md); this page is the grammar and
+the conventions.
+
+## Registers
+
+| Group | Names | Notes |
+|---|---|---|
+| General purpose, 64-bit | `AX` `BX` `CX` `DX` `SI` `DI` `BP` `SP` `R8` to `R15` | bare names, no prefix |
+| Sub-registers | `AL` `CL` `DL` `BL` `AH` family; `R8B` `R8W` `R8D` for the byte, word and double word of `R8` | width rides the mnemonic as well |
+| Vector | `X0` to `X15` (128-bit), `Y0` to `Y15` (256-bit), `Z0` to `Z31` (512-bit) | SSE, AVX and AVX-512 |
+| Mask | `K0` to `K7` | AVX-512 opmask |
+| System | `TLS` | the thread pointer, see below |
+
+Roles the calling convention fixes, which assembly must respect and can rely
+on:
+
+- `SP` is the hardware stack pointer; the virtual frame pointer of the
+  common language is the pseudo-register SP of OPERANDS.md, a different
+  spelling with a different meaning.
+- `BP` is callee-save. The assembler inserts the save and restore whenever
+  the function has a non-zero frame, so using BP as a general register
+  interferes with sampling profilers that walk the frame chain.
+- `R14` holds `g`, the goroutine pointer, in the register ABI; `RDX` holds
+  the closure context; `R12` and `R13` are the register ABI's scratch pair
+  and `R15` its GOT temporary; `X15` is the zeroing register the compiler
+  uses. An ABI0 assembly function called from Go sees none of these live
+  across the call, but runtime assembly reads them directly.
+- The legacy spellings for the goroutine pointer are the macros of
+  `runtime/go_tls.h`: `get_tls(r)` expands to `MOVQ TLS, r` and `g(r)` to
+  `0(r)(TLS*1)`, the segment base riding the index field.
+
+## Addressing
+
+The common forms of OPERANDS.md, with the amd64 specifics:
+
+```text
+offset(base)               MOVQ 16(BX), AX
+offset(base)(index*scale)  MOVL foo+32(SP)(R9*8), CX
+scale is 1, 2, 4 or 8
+name±offset(SB)            MOVQ ·table(SB), CX
+```
+
+- Global references assemble as absolute addresses and produce R_ADDR
+  relocations; branch targets produce R_PCREL.
+- Vector indexed memory, the VSIB form with an X, Y or Z register in the
+  index position, exists for the gather and scatter families.
+- There are no segment overrides in source; the one segment-flavoured form
+  is the TLS base in the index field shown above.
+
+## The frame and the split check
+
+The assembler manages the frame, not the programmer:
+
+- It inserts the `BP` save and restore for any non-zero frame.
+- It inserts the stack-split check for any function that is not NoSplit:
+  the check compares SP against the guard, and on exhaustion calls
+  `runtime.morestack_noctxt`. Frames at or below 128 bytes, StackSmall, use
+  the small compare; frames at or below 4096 bytes, StackBig, use the
+  adjusted form; larger frames compare in two steps.
+- On amd64 the assembler marks a function NoSplit itself when the frame is
+  under StackSmall and the body calls nothing that needs stack: such a
+  function carries the NoSplit flag in the object without the source ever
+  writing NOSPLIT.
+
+Results and arguments are stack-only in ABI0: the caller's frame carries
+them at FP offsets, per the Go prototype.
+
+## Instructions
+
+The inventory counts 1654 recognised mnemonics today, of which the encoder
+emits 1113; both numbers are generated in the appendix, and the gap is the
+encoder backlog that `gasm audit-instructions` measures. The families:
+
+- **Integer base.** The ALU and move set with width suffixes, `MOVB`,
+  `MOVW`, `MOVL`, `MOVQ`; the extension moves `MOVBLZX`, `MOVWLSX`,
+  `MOVLQSX` and their siblings, which the compiler's output leans on;
+  `LEA`; `PUSH` and `POP`; the shifts and rotates; the bit operations `BT`
+  through `BTC`, `BSF`, `BSR`, `LZCNT`, `TZCNT`, `POPCNT`, `BSWAP`; the
+  string primitives `MOVS` and `STOS`.
+- **Exchange and atomics.** `XCHG`, `CMPXCHG`, `XADD`; the extended-carry
+  pair `ADCX` and `ADOX`; `CRC32`.
+- **Scalar floating point.** The SSE2 scalar moves and arithmetic
+  (`MOVSD`, `MOVSS`, `ADDSD`, and the `CVT` family). Floating-point
+  immediates are not encodable on this target, so the assembler
+  materialises them: the constant lands in a synthesised read-only pool,
+  and a positive zero collapses to `XORPS` of the register with itself,
+  exactly as the toolchain does.
+- **Legacy SIMD, SSE.** The `MOVO`, `MOVOU`, `MOVAPS` family and the packed
+  integer and floating operations, shuffles, lane extracts and inserts and
+  the imm8-controlled forms.
+- **VEX and EVEX.** The `V`-prefixed forms for 256 and 512-bit work,
+  opmask operations on `K0` to `K7`, gathers and scatters, and the
+  quad-register families 4FMAPS, 4FNMADD, 4VNNIW, VP4DPWSSD and VP4DPWSSDS,
+  whose register list rides the inverted V′VVV field. Mixing VEX and legacy
+  SSE in one loop pays the AVX-SSE transition penalty on every switch: keep
+  a loop in one dialect.
+- **Cryptographic and counting extensions.** AES-NI, SHA-1 and SHA-256,
+  PCLMULQDQ, GFNI.
+- **System.** `CPUID`, `RDTSC`, `SYSCALL`, the fences, `LDMXCSR` and
+  `STMXCSR`, the prefetch family.
+- **Pseudo-operations.** `BYTE`, `WORD`, `LONG`, `QUAD` lay raw bytes or
+  words into the stream for encodings the assembler does not know; `ADJSP`
+  adjusts the stack pointer; `DUFFCOPY` and `DUFFZERO` and `GETCALLERPC`
+  are compiler-side names the table recognises but an encoder need not
+  emit.
+
+A mnemonic the appendix lists with `gasm encodes: no` assembles nowhere:
+gasm reports it as an explicit error, never as wrong bytes, and the
+`unencodable-instruction` lint flags it at edit time.
+
+## Relocations
+
+The relocations an amd64 object carries, all specified in
+[GOOBJ.md](../GOOBJ.md): `R_ADDR` for absolute globals, `R_PCREL` for
+relative addresses, `R_CALL` for direct calls, `R_TLS_LE` and `R_TLS_IE` for
+thread local access and `R_GOTPCREL` for GOT relative sequences, plus
+`R_DWTXTADDR_U4` inside the DWARF records, which the assembler always
+emits in the four-byte flavour.

docs/asm/ARM64.md

@@ -0,0 +1,122 @@
+# ARM64
+
+Layer 1, target page. Verified against `go tool asm` of Go 1.27.1, against the
+toolchain's own arm64 assembler manual (`cmd/internal/obj/arm64/doc.go`) and
+against gasm's encoder, whose output is compared byte for byte with the
+toolchain's. The complete mnemonic inventory lives in the generated appendix
+[INSTRUCTIONS-ARM64.md](INSTRUCTIONS-ARM64.md).
+
+## Registers
+
+- General purpose: `R0` to `R30`, plus `ZR`, the zero register, and `RSP`,
+  the stack pointer. There is no R31: thirty-one names and ZR.
+- Floating-point and SIMD share one file written `Vn`; where an instruction
+  is scalar floating point the operand may be written `Fn` (`F0` to `F31`).
+- SVE register names (`Z0` to `Z31`, `P0` to `P15`) exist in the assembler's
+  tables.
+- Roles the convention fixes: `RSP` is the stack pointer, `R29` the frame
+  pointer, `R30` the link register, `R26` the closure context and `R27` the
+  assembler's scratch register. The goroutine pointer lives in `R28` and is
+  written `g` in source, its fields as `g_m(g)`, `g_sched(g)`; `R18` is the
+  platform-reserved register and the Go toolchain never addresses it.
+
+## Loads, stores and the width suffixes
+
+The MOV series is the load and store interface, with the width in the
+mnemonic rather than the register name:
+
+| Mnemonic | Machine instruction |
+|---|---|
+| `MOVD` | ldr, str, stur, 64-bit |
+| `MOVW` | ldrsw, str, stur, 32-bit sign extending |
+| `MOVWU` | ldr, 32-bit zero extending |
+| `MOVH` | ldrsh, strh, sturh |
+| `MOVHU` | ldrh |
+| `MOVB` | ldrsb, strb, sturb |
+| `MOVBU` | ldrb |
+
+Post-index and pre-index addressing take the `.P` and `.W` suffixes on the
+mnemonic: `MOVD.P -8(R10), R8` is `ldr x8, [x10],#-8`, and `MOVB.W
+16(R16), R10` is `ldrsb x10, [x16,#16]!`.
+
+## Addressing
+
+```text
+imm(Rn|RSP)            28(R17)
+(Rn|RSP)               (R22)
+(Rn)(Rm)               (R27)(R23)
+(Rn)(Rm<<scale)        (R4)(R12<<2)
+(Rn)(Rm.UXTW<<3)       extended and shifted index
+(Rt1, Rt2)             register pair for LDP, STP and the exclusive pair forms
+```
+
+Branch targets are labels, `(R3)` for indirect, `name(SB)` for static.
+
+## Operand order and the special forms
+
+Most instructions appear in left-to-right assignment order: `ADD R11,
+RSP, R25` computes into R25. The exceptions the toolchain's manual lists,
+each with its own order:
+
+- stores and `CBZ`, `CBNZ` keep the GNU order: `MOVD R29, 384(R19)`.
+- The multiply-accumulate family `MADD`, `MSUB`, `SMADDL` and friends are
+  `<Rm>, <Ra>, <Rn>, <Rd>`.
+- The scalar FMA family `FMADDD` and friends are `<Fm>, <Fa>, <Fn>, <Fd>`.
+- The bitfield family `BFI`, `BFXIL`, `SBFIZ`, `SBFX`, `UBFIZ`, `UBFX` is
+  `$<lsb>, <Rn>, $<width>, <Rd>`.
+- The conditional compare and select families carry the condition as the
+  **first** operand: `CSEL GT, R0, R19, R1`, `CCMP MI, R22, $12, $13`,
+  `FCCMPD AL, F8, F26, $0`.
+- The exclusive stores are `<Rf>, (<Rn>), <Rs>` with the status register
+  last: `STLXR ZR, (R15), R16`.
+- `TBZ` and `TBNZ` are `$<imm>, <Rt>, <label>`.
+
+Shifted and extended register operands ride the register: `R19>>30`,
+`R26->24` for arithmetic right shift, `@>` for rotate, and the extend forms
+`R19.UXTB<<4`, `R14.SXTX` with extend operators UXTB, UXTH, UXTW, UXTX,
+SXTB, SXTH, SXTW, SXTX.
+
+## Conditions, branches and names
+
+- Conditions ride the branch mnemonic: `B.EQ`, or the canonical
+  per-condition names such as `BEQ`. Both spellings exist; the canonical
+  names are what the generated inventory lists.
+- `br` is `JMP` and `blr` is `CALL` in this dialect; indirect branches are
+  `JMP (R3)` and `CALL (R17)`.
+- `NOP` is a zero-width pseudo-instruction; the hardware nop is `NOOP`,
+  an alias of `HINT $0`.
+- `umov` is written as `VMOV`.
+
+## Constants
+
+- A 16-bit immediate optionally shifted: `MOVK $(10<<32), R20`, with
+  `MOVZ`, `MOVN` and their W variants; a zero shift is rejected by the
+  assembler.
+- Large integer constants: `MOV` materialises any 64-bit constant, the
+  closest-instruction way.
+- Vector constants: `VMOVS`, `VMOVD` and `VMOVQ`, the last taking two
+  64-bit halves for a 128-bit value:
+  `VMOVQ $0x1122334455667788, $0x99aabbccddeeff00, V2`.
+
+## SIMD
+
+Floating-point and SIMD instructions mostly carry a `V` prefix
+(`VADD`, `VFMLA`), the cryptographic extensions (`AESD`, `SHA256H`) and the
+scalar floating-point instructions being the exceptions. Operands carry an
+arrangement suffix, `V5.H8`, and structure loads and stores use bracket
+lists, `[V21.B16]`, with element selection as `V9.S[1]`.
+
+## Alignment
+
+`PCALIGN $n` pads to a power-of-two boundary between 8 and 2048 and also
+raises the function's alignment to the coarsest boundary any of its PCALIGN
+directives asks for. Functions default to 16-byte alignment on this target.
+
+## Relocations
+
+`R_ADDRARM64` for the adrp-plus-add pair, `R_ARM64_PCREL` and the
+`R_ARM64_PCREL_LDST` family for PC relative addressing, `R_ARM64_LDST` for
+the load and store immediates, `R_ARM64_GOTPCREL` and `R_ARM64_GOT` for the
+GOT, `R_ARM64_TLS_LE` and `R_ARM64_TLS_IE` for thread local storage and
+`R_CALLARM64` for direct calls, all specified in
+[GOOBJ.md](../GOOBJ.md).

docs/asm/DIRECTIVES.md

@@ -0,0 +1,148 @@
+# Directives: TEXT, DATA, GLOBL and the annotations
+
+Layer 1, the common language, with the flag vocabulary both layers share.
+Verified against `go tool asm` of Go 1.27.1, against the shipped headers
+`textflag.h` and `funcdata.h` in `$GOROOT/pkg/include`, and against gasm's
+parser. Where gasm extends a directive, the extension says so and is marked.
+
+Six directives exist. Three define things: TEXT, DATA, GLOBL. Three
+annotate: FUNCDATA, PCDATA, PCALIGN.
+
+## TEXT
+
+```text
+// func Add(a, b int64) int64
+TEXT ·Add(SB), NOSPLIT, $0-24
+	...instructions...
+	RET
+```
+
+```text
+TEXT symbol(SB), [flags,] $framesize[-argsize]
+```
+
+- The symbol is an `·Name(SB)` reference into the current package, or a
+  fully qualified name.
+- The optional flag argument is a constant expression, normally an OR of the
+  names from `textflag.h`, the table below. Without `#include "textflag.h"`
+  the names are not macros and the assembler reports the misleading error
+  `illegal or missing addressing mode for symbol NOSPLIT`: include the
+  header first.
+- `$framesize-argsize` is two constants, not a subtraction: the local frame
+  size in bytes, and the caller's argument area in bytes. The argument size
+  may be omitted entirely, `$16`, which marks the argument size unknown
+  (0x80000000 in the object, the value of `ArgsSizeUnknown` from
+  `funcdata.h`); a frame size may be negative only in the generated ABI
+  wrappers.
+- A function whose last instruction is not a branch cannot fall through into
+  the next TEXT: the toolchain appends a jump to itself, so end functions
+  with `RET` deliberately.
+- One TEXT per symbol; redeclaring is an error. The TEXT line also fixes the
+  function's source line for traceback: it is the line number that pcln
+  reports for the function's start.
+
+The framesize and argsize fields do real work: the framesize drives the
+stack-split preamble (RUNTIME.md carries the contract), and both travel into
+the FuncInfo record of the object (GOOBJ.md carries its layout).
+
+### The flag table
+
+Values from `textflag.h`, in agreement with `cmd/internal/obj/textflag.go`:
+
+| Name | Value | Applies to | Meaning |
+|---|---|---|---|
+| NOPROF | 1 | both | do not profile; deprecated |
+| DUPOK | 2 | both | the linker may keep one of several duplicates |
+| NOSPLIT | 4 | TEXT | no stack-split preamble |
+| RODATA | 8 | data | put the data in a read-only section |
+| NOPTR | 16 | data | the data contains no pointers |
+| WRAPPER | 32 | TEXT | a wrapper; must not disable `recover` |
+| NEEDCTXT | 64 | TEXT | a closure consuming the context register |
+| TLSBSS | 256 | data | a thread local word in BSS |
+| NOFRAME | 512 | TEXT | no frame setup; only valid with a frame size of 0 |
+| REFLECTMETHOD | 1024 | TEXT | the function calls `reflect.Type.Method` or `MethodByName` |
+| TOPFRAME | 2048 | TEXT | the outermost frame; unwinders stop here |
+| ABIWRAPPER | 4096 | TEXT | an ABI transition wrapper |
+
+Rules with teeth:
+
+- `NOSPLIT` removes the split check, so the frame plus everything the
+  function calls must fit in the stack segment that remains. It exists to
+  protect the splitting code itself; reaching for it to save two instructions
+  is how stack overflows corrupt memory. On amd64 the assembler additionally
+  marks small leaf functions NoSplit itself and omits the check, so the
+  absence of the preamble is not proof the flag was written.
+- A TEXT whose symbol is declared `ABIInternal` must carry NOSPLIT: the
+  assembler rejects it otherwise, because it cannot generate
+  the split path for a register-ABI function.
+- `RODATA` implies NOPTR for the garbage collector.
+
+## DATA
+
+```text
+DATA ·table+0(SB)/8, $0x0102030405060708
+DATA ·msg+0(SB)/14, $"hello, world\n"
+GLOBL ·msg(SB), RODATA, $14
+```
+
+```text
+DATA symbol+offset(SB)/width, value
+```
+
+- `width` is exactly 1, 2, 4 or 8: the initialiser is written into the data
+  image at `symbol+offset` in that many bytes.
+- The value is an integer or character constant of the width, or a string
+  literal whose byte length equals the width exactly; escapes count. Long
+  data is written as successive DATA lines at increasing offsets; bytes the
+  directives never name are zero.
+- Every symbol initialised with DATA ends with a GLOBL line declaring its
+  total size, after all of its DATA lines.
+
+A symbol containing pointers cannot be defined in assembly, because the
+collector cannot see into it: define it in Go and refer to it by name. As a
+rule, data that is not read-only belongs in Go.
+
+Extension, gasm only: a DATA initialiser may name a symbol,
+`DATA ·fn+0(SB)/8, $·handler(SB)`, which gasm lays down as an absolute
+relocation on that field. The toolchain offers no ground truth for this
+form; gasm's behaviour is verified by linking and execution.
+
+## GLOBL
+
+```text
+GLOBL symbol(SB), [flags,] $size
+```
+
+Declares the symbol global with its total size in bytes. The useful flags
+are RODATA, NOPTR, DUPOK and TLSBSS from the table above. Uninitialised
+bytes are zero, which makes GLOBL with no DATA the language's BSS.
+
+## FUNCDATA and PCDATA
+
+```text
+FUNCDATA $functypeid, symbol(SB)
+PCDATA $pctypeid, $value
+```
+
+The compiler's annotations for the garbage collector and traceback, named by
+the ids in `funcdata.h`: FUNCDATA 0 to 7 (args pointer maps, locals pointer
+maps, stack objects, inline tree, open-coded defer info, argument info,
+argument liveness, wrap info), PCDATA 0 to 4 (unsafe point, stack map index,
+inline tree index, argument liveness index, panic bounds). Assembly code
+normally reaches them only through the macro forms in `funcdata.h`, which
+RUNTIME.md explains. Outside the macros, hand-written PCDATA is meaningless:
+the values are pc-value tables the compiler builds from its own view of the
+program.
+
+## PCALIGN
+
+```text
+PCALIGN $32
+```
+
+Pads the code so that the next instruction lands on the given boundary,
+which must be a power of two and at least the target's instruction
+alignment. Supported on amd64, arm64, ppc64, loong64 and riscv64. The
+padding instructions are the target's NOP encoding, so the bytes between
+functions differ from what the instruction stream alone would produce, which
+matters to anyone comparing encodings byte for byte.