[PATCH v3 0/11] OpenRISC orfpx64a32 and openrisc spec 1.3 support
Stafford Horne
shorne@gmail.com
Sat Jun 8 21:32:00 GMT 2019
Hello,
Its been a while since v2, but we have been busy changing direction a bit.
Instead of just adding the orfpx64a32 extension support we have created a new
architecture specification [0] and this series includes some of those changes.
Changes from v2:
- Use explicit register pairs in orfpx64a32 extension instead of implicit
suggestioned by rth.
- Add unordered FPU comparison instructions.
- Add test for the previously added l.adrp instruction.
- A few non material cgen comment/symbol updates.
Changes from v1:
- Rebased
- Note Andrey is working on doing his FSF copyright assignment for binutils,
sim
The orfpx64a32 extension patches are to the OpenRISC assembler and simulator to
support 64-bit floating point operations on 32-bit cores using register pairs,
see orfpx64a32 [1].
The original patches were written by Andrey Bacherov and I have made some
updates to get the simulator working and to match ABI updates for the final
architecture 1.3 specification.
Example of orfpx64a32 operations:
lf.add.d r3,r4, r5,r6, r7,r8 ; {r3,r4} <= {r5,r6} + {r7,r8}
lf.add.d r16,r18, r20,r22, r24,r26 ; {r16,r18} <= {r20,r22} + {r24,r26}
Note, on 64-bit architectures the instructions will be written with 64-bit
registers.
lf.add.d r3, r5, r7 ; r3 <= r5 + r7
lf.add.d r16, r20 r24 ; r16 <= r20 + r24
The new unordered comparison instructions allow the OpenRISC FPU to do
comparisons where input may include NaNs [2]
These binutil patches have been used along with the GCC FPU patches [3] on the
OpenRISC GCC 9 toolchain to test single and double precision floating point
support including unordered comparisons. The main 'real' implementation of this
hardware is Andrey's or1k_marocchino [4] core implementation which we have been
using along with simulators for verification.
This whole patch series can be found on my github repo [5] as well.
[0] https://github.com/openrisc/doc/blob/master/openrisc-arch-1.3-rev1.pdf
[1] https://openrisc.io/proposals/orfpx64a32
[2] https://openrisc.io/proposals/lfsf
[3] git@github.com:stffrdhrn/gcc.git or1k-fpu-2
[4] https://github.com/openrisc/or1k_marocchino
[5] git@github.com:stffrdhrn/binutils-gdb.git orfpx64a32-3
-Stafford
Stafford Horne (11):
cpu/or1k: Add support for orfp64a32 spec
cpu/or1k: Define unordered comparisons
cpu/or1k: Document no branch delay slot architectures and l.adrp
cpu/or1k: Update fpu compare symbols to imply set flag
opcodes/or1k: Regenerate opcodes
sim/or1k: Regenerate sim
sim/common: Wire in df/di conversion
sim/common: wire up new unordered comparisons
sim/testsuite/or1k: Add test for 64-bit fpu operations
sim/testsuite/or1k: Add test case for l.adrp instruction
sim/testsuite/or1k: Add tests for unordered compares
cpu/or1k.cpu | 21 +-
cpu/or1k.opc | 92 ++
cpu/or1kcommon.cpu | 113 +-
cpu/or1korbis.cpu | 3 +-
cpu/or1korfpx.cpu | 214 ++-
opcodes/or1k-asm.c | 72 +-
opcodes/or1k-desc.c | 309 +++-
opcodes/or1k-desc.h | 341 ++---
opcodes/or1k-dis.c | 43 +-
opcodes/or1k-ibld.c | 234 ++-
opcodes/or1k-opc.c | 319 +++-
opcodes/or1k-opc.h | 39 +-
opcodes/or1k-opinst.c | 100 +-
sim/common/cgen-accfp.c | 49 +
sim/common/cgen-fpu.h | 2 +
sim/or1k/arch.c | 2 +-
sim/or1k/arch.h | 2 +-
sim/or1k/cpu.c | 78 +-
sim/or1k/cpu.h | 174 ++-
sim/or1k/cpuall.h | 2 +-
sim/or1k/decode.c | 568 ++++++--
sim/or1k/decode.h | 21 +-
sim/or1k/model.c | 1456 ++++++++++++++++---
sim/or1k/sem-switch.c | 602 +++++++-
sim/or1k/sem.c | 672 ++++++++-
sim/testsuite/sim/or1k/adrp.S | 73 +
sim/testsuite/sim/or1k/fpu-unordered.S | 97 ++
sim/testsuite/sim/or1k/fpu64a32-unordered.S | 100 ++
sim/testsuite/sim/or1k/fpu64a32.S | 172 +++
29 files changed, 5253 insertions(+), 717 deletions(-)
create mode 100644 sim/testsuite/sim/or1k/adrp.S
create mode 100644 sim/testsuite/sim/or1k/fpu-unordered.S
create mode 100644 sim/testsuite/sim/or1k/fpu64a32-unordered.S
create mode 100644 sim/testsuite/sim/or1k/fpu64a32.S
--
2.21.0
More information about the Gdb-patches
mailing list