Till then we had separate stage to write results of
an operations to registers. Now operations will write
result immediately to a destination register. Registers
will be saved to a temporary location if following
operations in a bundle must read them.
The change elliminates additional branches for
conditional operations in the write stage and reduces
data movements.
Signed-off-by: Denis Drakhnya <numas13@gmail.com>
Fix time fields in stat64.
Restore and save a breakpoint frame in a signal
handler. This will fix a segfault if a signal will
be raised while a cpu is in a breakpoint.
Signed-off-by: Denis Drakhnya <numas13@gmail.com>
v5+ has different procedure stack layout. Reading
the stack through the acees_hw_stacks syscall
requires to shuffle registers for a backward
compatibility with previous versions.
Signed-off-by: Denis Drakhnya <numas13@gmail.com>
The TCG translation was splitted into multiple files and
it was painful because it has a lot of cross references.
Reorg FX helpers.
Signed-off-by: Denis Drakhnya <numas13@gmail.com>
Previously was used two separate arrays for low and
high halves of a register. Now we can pass direct
pointer to a register into helpers instead of copying
the halves into a temporary buffer and pass pointer to
that buffer.
Signed-off-by: Denis Drakhnya <numas13@gmail.com>
QEMU has cpu_exec_step_atomic function to run
one instruction in an exclusive mode but we need
to execute multiple instructions in that mode.
I don't want to modify the function.
The solution provides a correct execution of
atomic operations like on real CPU but needs
more testing to prove that it works correctly
in all cases.
Signed-off-by: Denis Drakhnya <numas13@gmail.com>
objdump prints incorrect argument sizes of some
integer vector operations from the first version
of ISA.
Signed-off-by: Denis Drakhnya <numas13@gmail.com>
Dynamically expanding the stack has not yet been
implemented, but sometimes applications require
larger stacks.
Signed-off-by: Denis Drakhnya <numas13@gmail.com>
DAM has not been yet implemented. We always generate
a jump to a fixing code. All these messages are redundant.
Signed-off-by: Denis Drakhnya <numas13@gmail.com>
We cannot just change the IP address in E2K to restart
the instruction because the same register can be read
and written in the same instruction (and many more
situations), but we can create a fake syscall instruction
and restart it.
Signed-off-by: Denis Drakhnya <numas13@gmail.com>
