Now addressing the complex vector op issue. I now expose TCGv_vec to target front-ends, but opaque wrt the vector size. One can thus compose vector operations, as demonstrated in target/arm/.
The actual host vector length now becomes an argument to the *_vec opcodes. It's a little awkward, but does prevent an explosion of opcode values. All R-b dropped because all patches rewritten or heavily modified. Whacha think? r~ Richard Henderson (6): tcg: Add types and operations for host vectors tcg: Add vector expanders target/arm: Align vector registers target/arm: Use vector infrastructure for aa64 add/sub/logic tcg/i386: Add vector operations tcg/aarch64: Add vector operations Makefile.target | 2 +- accel/tcg/tcg-runtime.h | 24 ++ target/arm/cpu.h | 2 +- tcg/aarch64/tcg-target.h | 20 +- tcg/i386/tcg-target.h | 36 +- tcg/tcg-gvec-desc.h | 49 +++ tcg/tcg-op-gvec.h | 143 ++++++++ tcg/tcg-op.h | 26 ++ tcg/tcg-opc.h | 37 ++ tcg/tcg.h | 34 ++ accel/tcg/tcg-runtime-gvec.c | 255 +++++++++++++ target/arm/translate-a64.c | 216 +++++++---- tcg/aarch64/tcg-target.inc.c | 340 ++++++++++++++--- tcg/i386/tcg-target.inc.c | 423 ++++++++++++++++++--- tcg/tcg-op-gvec.c | 853 +++++++++++++++++++++++++++++++++++++++++++ tcg/tcg-op.c | 234 ++++++++++++ tcg/tcg.c | 77 +++- accel/tcg/Makefile.objs | 2 +- tcg/README | 46 +++ 19 files changed, 2651 insertions(+), 168 deletions(-) create mode 100644 tcg/tcg-gvec-desc.h create mode 100644 tcg/tcg-op-gvec.h create mode 100644 accel/tcg/tcg-runtime-gvec.c create mode 100644 tcg/tcg-op-gvec.c -- 2.13.5