We currently search both the root and the tcg/ directories for tcg
files:
$ git grep '#include "tcg/' | wc -l
28
$ git grep '#include "tcg[^/]' | wc -l
94
To simplify the preprocessor search path, unify by expliciting the
tcg/ directory.
Patch created mechanically by running:
$ for x in \
tcg.h tcg-mo.h tcg-op.h tcg-opc.h \
tcg-op-gvec.h tcg-gvec-desc.h; do \
sed -i "s,#include \"$x\",#include \"tcg/$x\"," \
$(git grep -l "#include \"$x\""); \
done
Acked-by: David Gibson <david@gibson.dropbear.id.au> (ppc parts)
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Stefan Weil <sw@weilnetz.de>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20200101112303.20724-2-philmd@redhat.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
These are now generated by decodetree itself.
Tested-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Acked-by: Palmer Dabbelt <palmer@sifive.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
fence.i has been split out of the base ISA as part of the ratification
process. This patch adds a Zifencei argument, which disables the
fence.i instruction.
Signed-off-by: Palmer Dabbelt <palmer@sifive.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
The tcg_gen_fooi_tl functions have some immediate constant
folding built in, which match up with some of the riscv asm
builtin macros, like mv and not.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Palmer Dabbelt <palmer@sifive.com>
Signed-off-by: Palmer Dabbelt <palmer@sifive.com>
This eliminates all functions in insn_trans/trans_rvc.inc.c,
so the entire file can be removed.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Palmer Dabbelt <palmer@sifive.com>
Signed-off-by: Palmer Dabbelt <palmer@sifive.com>
Special handling for IMM==0 is the only difference between
RVC shifti and RVI shifti. This can be handled with !function.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Palmer Dabbelt <palmer@sifive.com>
Signed-off-by: Palmer Dabbelt <palmer@sifive.com>
The generated functions are only used within translate.c
and do not need to be global, or declared.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Palmer Dabbelt <palmer@sifive.com>
Signed-off-by: Palmer Dabbelt <palmer@sifive.com>
This patch introduces wrappers around the tcg_gen_exit_tb() and
tcg_gen_lookup_and_goto_ptr() functions that handle single stepping,
i.e. call gen_exception_debug() when single stepping is enabled.
Theses functions are then used instead of the originals, bringing single
stepping handling in places where it was previously ignored such as jalr
and system branch instructions (ecall, mret, sret, etc.).
Signed-off-by: Fabien Chouteau <chouteau@adacore.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Signed-off-by: Palmer Dabbelt <palmer@sifive.com>
In order to handle TB's that translate to too much code, we
need to place the control of the length of the translation
in the hands of the code gen master loop.
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
While running the GCC test suite against 4.0.0-rc0, Kito found a
regression introduced by the decodetree conversion that caused divuw and
remuw to sign-extend their inputs. The ISA manual says they are
supposed to be zero extended:
DIVW and DIVUW instructions are only valid for RV64, and divide the
lower 32 bits of rs1 by the lower 32 bits of rs2, treating them as
signed and unsigned integers respectively, placing the 32-bit
quotient in rd, sign-extended to 64 bits. REMW and REMUW
instructions are only valid for RV64, and provide the corresponding
signed and unsigned remainder operations respectively. Both REMW
and REMUW always sign-extend the 32-bit result to 64 bits, including
on a divide by zero.
Here's Kito's reduced test case from the GCC test suite
unsigned calc_mp(unsigned mod)
{
unsigned a,b,c;
c=-1;
a=c/mod;
b=0-a*mod;
if (b > mod) { a += 1; b-=mod; }
return b;
}
int main(int argc, char *argv[])
{
unsigned x = 1234;
unsigned y = calc_mp(x);
if ((sizeof (y) == 4 && y != 680)
|| (sizeof (y) == 2 && y != 134))
abort ();
exit (0);
}
I haven't done any other testing on this, but it does fix the test case.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Palmer Dabbelt <palmer@sifive.com>
decodetree handles all instructions now so the fallback is not necessary
anymore.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Bastian Koppelmann <kbastian@mail.uni-paderborn.de>
Signed-off-by: Peer Adelt <peer.adelt@hni.uni-paderborn.de>
with all 16 bit insns moved to decodetree no path is falling back to
gen_system(), so we can remove it.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Bastian Koppelmann <kbastian@mail.uni-paderborn.de>
Signed-off-by: Peer Adelt <peer.adelt@hni.uni-paderborn.de>
manual decoding in gen_arith() is not necessary with decodetree. For now
the function is called trans_arith as the original gen_arith still
exists. The former will be renamed to gen_arith as soon as the old
gen_arith can be removed.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Bastian Koppelmann <kbastian@mail.uni-paderborn.de>
Signed-off-by: Peer Adelt <peer.adelt@hni.uni-paderborn.de>
gen_arith_imm() does a lot of decoding manually, which was hard to read
in case of the shift instructions and is not necessary anymore with
decodetree.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Bastian Koppelmann <kbastian@mail.uni-paderborn.de>
Signed-off-by: Peer Adelt <peer.adelt@hni.uni-paderborn.de>
With decodetree we don't need to convert RISC-V opcodes into to MemOps
as the old gen_store() did.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Bastian Koppelmann <kbastian@mail.uni-paderborn.de>
Signed-off-by: Peer Adelt <peer.adelt@hni.uni-paderborn.de>
With decodetree we don't need to convert RISC-V opcodes into to MemOps
as the old gen_load() did.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Bastian Koppelmann <kbastian@mail.uni-paderborn.de>
Signed-off-by: Peer Adelt <peer.adelt@hni.uni-paderborn.de>
We now utilizes argument-sets of decodetree such that no manual
decoding is necessary.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Bastian Koppelmann <kbastian@mail.uni-paderborn.de>
Signed-off-by: Peer Adelt <peer.adelt@hni.uni-paderborn.de>
trans_jalr() is the only caller, so move the code into trans_jalr().
Acked-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Bastian Koppelmann <kbastian@mail.uni-paderborn.de>
Signed-off-by: Peer Adelt <peer.adelt@hni.uni-paderborn.de>
we cannot remove the call to gen_arith() in decode_RV32_64G() since it
is used to translate multiply instructions.
Acked-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Bastian Koppelmann <kbastian@mail.uni-paderborn.de>
Signed-off-by: Peer Adelt <peer.adelt@hni.uni-paderborn.de>
this splits the 64-bit only instructions into its own decode file such
that we generate the decoder for these instructions only for the RISC-V
64 bit target.
Acked-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Bastian Koppelmann <kbastian@mail.uni-paderborn.de>
Signed-off-by: Peer Adelt <peer.adelt@hni.uni-paderborn.de>
for now only LUI & AUIPC are decoded and translated. If decodetree fails, we
fall back to the old decoder.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Acked-by: Alistair Francis <alistair.francis@wdc.com>
Signed-off-by: Bastian Koppelmann <kbastian@mail.uni-paderborn.de>
Signed-off-by: Peer Adelt <peer.adelt@hni.uni-paderborn.de>
Add misa checks for M, A, F and D extensions and if they are
not present generate illegal instructions. This improves
emulation accurary for harts with a limited set of extensions.
Signed-off-by: Michael Clark <mjc@sifive.com>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Palmer Dabbelt <palmer@sifive.com>
gen methods should access state from DisasContext. Add misa
field to the DisasContext struct and remove CPURISCVState
argument from all gen methods.
Signed-off-by: Michael Clark <mjc@sifive.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
Signed-off-by: Palmer Dabbelt <palmer@sifive.com>
The gen methods should access state from DisasContext. Add priv_ver
field to the DisasContext struct.
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Palmer Dabbelt <palmer@sifive.com>
Modifed from Richard Henderson's patch [1] to integrate
with the new control and status register implementation.
[1] https://lists.nongnu.org/archive/html/qemu-devel/2018-03/msg07034.html
Note: the f* CSRs already mark mstatus.FS dirty using
env->mstatus |= mstatus.FS so the bug in the first
spin of this patch has been fixed in a prior commit.
Signed-off-by: Michael Clark <mjc@sifive.com>
Reviewed-by: Michael Clark <mjc@sifive.com>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
Co-authored-by: Richard Henderson <richard.henderson@linaro.org>
Co-authored-by: Michael Clark <mjc@sifive.com>
Signed-off-by: Palmer Dabbelt <palmer@sifive.com>
Signed-off-by: Michael Clark <mjc@sifive.com>
Reviewed-by: Michael Clark <mjc@sifive.com>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
Signed-off-by: Palmer Dabbelt <palmer@sifive.com>
Our current fence implementation ignores fences for the user-only
configurations. This is incorrect but unlikely to manifest: it requires
multi-threaded user-only code that takes advantage of the weakness in
the host's memory model and can be inlined by TCG.
This patch simply treats fences the same way for all our emulators.
I've given it to testing as I don't want to construct a test that would
actually trigger the failure.
Our fence implementation has an additional deficiency where we map all
RISC-V fences to full fences. Now that we have a formal memory model
for RISC-V we can start to take advantage of the strength bits on our
fence instructions. This requires a bit more though, so I'm going to
split it out because the implementation is still correct without taking
advantage of these weaker fences.
Thanks to Richard Henderson for pointing out both of the issues.
Signed-off-by: Palmer Dabbelt <palmer@sifive.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
sfence.vm has been replaced in priv v1.10 spec by sfence.vma.
Reported-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Bastian Koppelmann <kbastian@mail.uni-paderborn.de>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Palmer Dabbelt <palmer@sifive.com>
Performance impact of this and the previous commits, measured with
the very-easy-to-cross-compile rv8-bench:
https://github.com/rv8-io/rv8-bench
Host: Intel(R) Core(TM) i7-4790K CPU @ 4.00GHz
- Key:
before: master
after1,2,3: the 3 commits in this series (i.e. 3 is this commit)
- User-mode:
bench before after1 after2 after3 final speedup
---------------------------------------------------------
aes 1.12s 1.12s 1.10s 1.00s 1.12
bigint 0.78s 0.78s 0.78s 0.78s 1
dhrystone 0.96s 0.97s 0.49s 0.49s 1.9591837
miniz 1.94s 1.94s 1.88s 1.86s 1.0430108
norx 0.51s 0.51s 0.49s 0.48s 1.0625
primes 0.85s 0.85s 0.84s 0.84s 1.0119048
qsort 4.87s 4.88s 1.86s 1.86s 2.6182796
sha512 0.76s 0.77s 0.64s 0.64s 1.1875
(after1 only applies to softmmu, so no surprises here)
- Full-system (fedora):
bench before after1 after2 after3 final speedup
---------------------------------------------------------
aes 2.68s 2.54s 2.60s 2.34s 1.1452991
bigint 1.61s 1.56s 1.55s 1.64s 0.98170732
dhrystone 1.78s 1.67s 1.25s 1.24s 1.4354839
miniz 3.53s 3.35s 3.28s 3.35s 1.0537313
norx 1.13s 1.09s 1.07s 1.06s 1.0660377
primes 15.37s 15.41s 15.20s 15.37s 1
qsort 7.20s 6.71s 3.85s 3.96s 1.8181818
sha512 1.07s 1.04s 0.90s 0.90s 1.1888889
SoftMMU slows things down, so the numbers are less sensitive.
Cross-page jumps improve things a little bit, though.
Note that I'm not showing here averages, just results from a
single run, so with primes there isn't much to worry about.
Signed-off-by: Emilio G. Cota <cota@braap.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
Philippe Mathieu-Daudé
hiroyuki.obinata
Emilio G. Cota
Richard Henderson
Palmer Dabbelt
Fabien Chouteau
Bastian Koppelmann
Michael Clark
Alistair Francis