* sky-vu1.c (vu1_io_write_register_window): Make CIA (pc) write
effective by updating more registers.
* sky-libvpe.c: Updated to match earlier VU state-change code.
* sky-vpe.h: Removed unused globals from declarations.
Recognize -GN and -relax.
(md_begin): Initialize gp size from -G switch.
(alpha_force_relocation): Always force if -relax.
(alpha_align): Take a new argument that will specify when to
emit an R_ALPHA_ALIGN relocation (though we don't do that now).
Change all callers. Emit nop alignment padding as nop+unop pair.
PKE tests run identically on SPARC/Solaris and x86/Linux.
* sky-pke.c (pke_io_{read,write}_buffer): Endianness fixes aka
"E-fixes" in register and FIFO read/writes.
(pke_code_{pkemscalf,pkemscal}): E-fixes in VU CIA setting.
(pke_code_{mpg,unpack}): E-fixes in VU memory & tracking updates.
(pke_code_direct): E-fixes in GPUIF FIFO stuffing.
* sky-pke.h (PKE_MEM_WRITE): E-fixes in trace file writing.
* sky-vu0.c (vu0_attach): Allocate micro/data memory with zalloc
to guarantee sufficient (16-byte) alignment.
* sky-vu1.c (vu1_attach): Ditto.
(vu1_io_read_register_window): *PARTIAL* E-fixes in register accesses.
* sky-libvpe.c (gif_write): E-fixes in GPUIF FIFO stuffing.
* sky-gpuif.c (gif_io_{read,write}_buffer): E-fixes in
register and FIFO read/writes.
* sky-dma.c (do_dma_transfer_tag): E-fixes in tag reading.
(md_shortopts): Add -O.
(md_parse_option): Recognize it.
(md_show_usage): Document it.
(md_assemble): Don't try to combine insns in parallel if !optimize.
mechanism is starting to subside.
* sky-pke.h (PKE_FLAG_INT_NOLOOP): Added device flag to indicate
presence of stalled & interrupted PKEcode.
* sky-pke.c (pke_issue): Added PKEcode interrupt bit handling.
(pke_flip_dbf): Changed double-buffering logic to match SCEI
clarification.
(pke_code_*): Added interrupt bit stalling clause.
(pke_code_pkems*): Added ITOP/ITOPS transmission code.
(pke_code_unpack): Added more careful logic for processing
overflows of VU data memory addresses.
Handle mips address vs vu address.
* dvp-opc.c (vif_operands): Use DVP_OPERAND_VU_ADDRESS.
(dma_operands): Use DVP_OPERAND_MIPS_ADDRESS.
({insert,extract}_dma_addr): Fix word ofset.
({insert,print}_gif_regs): Fix encode/decode.
(engine_resume_full): Keep accurate core profile data.
* cgen-utils.c (sim_disassemble_insn): Don't use
sim_core_read_aligned_N, it messes up profiling results.
