Skip to content

Latest commit

 

History

History
191 lines (152 loc) · 6.56 KB

CHANGELOG.md

File metadata and controls

191 lines (152 loc) · 6.56 KB
Raw

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog and this project adheres to Semantic Versioning.

[Unreleased]

[2.6.0] - 2022-11-09

Added

  • Re-added .tiny section support

Fixed

  • Removed backwards compatibility warnings when using xpulpv2 and mul extensions

[2.5.0] - 2021-10-30

Added

  • gcc: Add xfhalf, xfalthalf, xfhalfinx and xfhalthalfinx support. This means code generation in the backend, libgcc changes to make softfloat routines set the fcsr, itanium style name mangling rules for and some helper routines when truncating doubles.
  • gcc: Add xfhalf and xfalthalf compile test
  • binutils: Add xfhalfinx and xfalthalfinx support
  • gdb/sim: Add xfhalf and xfhalfinx support
  • gdb/sim: Add float and half float tests generated by patched TestFloat-3e
  • newlib: Add support for zfinx (POSIX fenv)

Changed

  • gcc, binutils: Conversion instructions between half float and other float modes are automatically inferred. Internally, they are stil handled separately to make changes such as selectively disabling them easier.

Fixed

  • binutils: Add fcsr alias instructions for zfinx
  • gdb/sim: NaN boxing for floats in double float registers
  • gdb/sim: dynamic rounding mode didn't work at all
  • gdb/sim: Add RISC-V NaN behavior (canonical NaN instead of complicated rules)
  • gcc: Various typos in tests
  • gcc: Hardware loops for the xcorev target are now disabled (due to wrong code generation)
  • gcc: Nested looping code now properly results in nested hardware loop instructions

[2.4.0] - 2021-07-27

Added

  • gcc, binutils, sim: Add support for zfinx and zdinx (rv32 and rv64). Use zfinx (replaces f) and zdinx (replaces d).
  • gcc, binutils: Add basic zfinx/zdinx tests
  • Added zfinx multilibs

Changed

  • sim: Now uses arch attribute from elf do determine supported extension

Fixed

  • gcc: Fix __builtin_pulp_ClusterId() and add test
  • gdb: backported fix for NULL pointer deref when using focus or winheight command (only happens when using a bleeding edge gcc)
  • sim: Fix postmod load/store when using the same register as destination and source.

[2.3.0] - 2021-07-19

Added

  • gcc: Add simd complex insn
  • gcc: Add pulp specific tests

Changed

  • gcc: Generalize simd md patterns
  • gcc: builtins are now available depending on enabled subextensions instead of pulp core version hackery.

Fixed

  • gcc: xpulpvect was not working as argument, but xpulpvectall was
  • gcc: Fix single standing xpulphwloop arch string

[2.2.0] - 2021-07-09

Added

  • gdb/sim: Add pulp vector insn
  • gdb/sim: Add bitrev insn
  • gdb/sim: Add testsuite

Fixed

  • gas: Add p.bitrev and simd complex to pulpv3 and corev
  • gdb/sim: Fix clip boundaries
  • gdb/sim: Improve error handling
  • gcc: V2QI moves
  • gcc: Use pv.pack instead of pv.pack.h

Changed

  • gdb/sim: Set default simulation model to corev
  • gcc: Enable SIMD support. Was functional, but disabled.

[2.1.3] - 2021-07-05

Added

  • Allow setting pkversion string (--with-pkgversion=ABC)

Changed

  • gdb, sim: Rebased binutils patches

[2.1.2] - 2021-07-03

Added

  • gcc: Tweaks for hardware loops (minimum size and alignment)

[2.1.1] - 2021-07-03

Fixed

  • gcc: PULP builtins fail during reload/lra pass

[2.1.0] - 2021-07-03

Added

  • pulp.md about new subextensions and extension groupings
  • core-v.rst about using cv32e40p with PULP extensions
  • gcc: Enable/Disable PULP extension with -m flags (see --help and pulp.md)
  • gcc: Updated multilib settings
  • gcc,binutils: Preliminary support for cv32e40p (xcorev)
  • binutils: smach custom float extension and extension strings
  • bintuils: Imported and patched corev assembly tests

Changed

  • gcc: Split PULP extensions into various subextensions. Check pulp.md.
  • gcc: Split PULP specific machine descriptions into separate files
  • binutils/gas: More warnings for hardware loops when used in a wrong way
  • binutils: Single location for PULP extension meta information making tedious copy-paste-sync superfluous

Fixed

  • gcc,binutils: Version parsing of arch string (upstream bug)
  • gcc: Code generation for __builtin_clrsb
  • binutils: Disassembly of ji immeadite of hardware loops

Removed

  • gcc: Various dead -m options
  • binutils: Dead code from custom dynamic linking

[2.0.1] - 2020-11-19

Changed

  • Using a extension with xpulpv2 is not an error anymore
  • Remove redundant code that forces abi to ilp32 when using xpulp

[2.0.0] - 2020-11-4

Added

  • gcc: port pulp patches to 9.2.0
  • binutils: port pulp patches to 9.2.0
  • gdb: port pulp patches to 9.2.0
  • gcc: Add support xpulpv2 except vector insn (not verified) and ind-ind-reg insn (can't get patterns to work)
  • gcc: Use new register allocator (lra) instead of the deprecated reload.
  • gcc: Properly gate pulp insns behind pulp arch flags. This makes sure when using standard RISC-V arch flags that we clearly don't use patched code paths.
  • gcc: Reintroduce default RISC-V builtins
  • gdb: Add xpulpv2 insn to gdb-simulator. Used as canonical simulator to run regression tests.
  • ld: Add full support for pulp relocations
  • as: Remap operand shorthands of insn to resolve collisions
  • as: Adapt table and glue logic to new insn class logic
  • objdump: Add support for parsing of riscv-attributes to determine arch. Useful when trying to disassemble binaries that use insn with overlapping encodings.
  • Add gitlabci flow

Changed

  • Most tools only accept lower case -march and -mabi arguments like upstream gcc does.
  • gcc: xpulpv2 will fail when combined with a extension. Previously the a extension was just silently stripped.
  • objdump: insn that were previously rendered as pulp insn (prefix p.) but actually weren't are now shown in their original form (e.g. mul instead of p.mul)

Fixed

  • gcc: Fix __attribute__ ((visibility ("hidden"))) logic
  • gcc: Dismantle mnosext. Broken semantics and fixes regression with wrong insn being emitted.
  • gcc: Disable Pulp_DP_Format_type support
  • gcc: Disable mdf (internal) flag
  • gcc: Remove the automatic (broken) linking to a file called riscv.ld in riscv/elf.h
  • gcc: Fix conflicts in constraint naming
  • gcc: Disable high part insn code path in gcc expand pass. Looks like a gcc bug.

Removed

  • ld: Remove lightweight dynamic linking (buggy and interfering with default linker code)
  • gcc: Remove PULP specific interrupt handler attributes. The code was contending with upstream interrupt handler support.