[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Qemu-riscv] [Qemu-devel] [RFC v1 15/23] riscv: tcg-target: Add bran
From: |
Alistair Francis |
Subject: |
Re: [Qemu-riscv] [Qemu-devel] [RFC v1 15/23] riscv: tcg-target: Add branch and jump instructions |
Date: |
Tue, 20 Nov 2018 15:49:29 -0800 |
On Fri, Nov 16, 2018 at 1:14 AM Richard Henderson
<address@hidden> wrote:
>
> On 11/15/18 11:36 PM, Alistair Francis wrote:
> > +static void tcg_out_brcond(TCGContext *s, TCGCond cond, TCGReg arg1,
> > + TCGReg arg2, TCGLabel *l)
> > +{
> > + RISCVInsn op = tcg_brcond_to_riscv[cond].op;
> > + bool swap = tcg_brcond_to_riscv[cond].swap;
> > +
> > + tcg_out_opc_branch(s, op, swap ? arg2 : arg1, swap ? arg1 : arg2, 0);
>
> You might want to tcg_debug_assert(op != 0) here.
>
> > + if (l->has_value) {
> > + reloc_sbimm12(s->code_ptr - 1, l->u.value_ptr);
>
> I'm concerned about the conditional branch range. +-4K isn't much to work
> with. The minimum we have for other hosts is +-32K.
>
> We have two options: (1) greatly reduce the max size of the TB for this host;
> (2) be prepared to emit a 2 insn sequence: conditional branch across
> unconditional branch, with forward branches that turn out to be small patched
> with a nop.
>
> FWIW, the first case would be done via modification of tcg_op_buf_full. You
> might have to go as low as 500 opcodes, I'm not sure.
How do we do option 2?
>
> > +static inline void tcg_out_goto(TCGContext *s, tcg_insn_unit *target)
> > +{
> > + ptrdiff_t offset = tcg_pcrel_diff(s, target);
> > + tcg_debug_assert(offset == sextract64(offset, 0, 26));
> > + tcg_out_opc_jump(s, OPC_JAL, TCG_REG_ZERO, offset);
> > +}
> > +
> > +static inline void tcg_out_goto_long(TCGContext *s, tcg_insn_unit *target)
> > +{
> > + ptrdiff_t offset = tcg_pcrel_diff(s, target);
> > +
> > + if (offset == sextract64(offset, 0, 26)) {
> > + tcg_out_opc_jump(s, OPC_JAL, TCG_REG_ZERO, offset);
> > + } else {
> > + tcg_out_movi(s, TCG_TYPE_I64, TCG_REG_TMP0, (intptr_t)target);
> > + tcg_out_opc_jump(s, OPC_JAL, TCG_REG_TMP0, 0);
> > + }
> > +}
>
> How are these to be used? I guess I'll find out...
It's called on a exit TB op. I'm not sure if that's correct though.
Alistair
>
> > +static void tcg_out_call_int(TCGContext *s, tcg_insn_unit *arg, bool tail)
> > +{
> > + TCGReg link = tail ? TCG_REG_ZERO : TCG_REG_RA;
> > + ptrdiff_t offset = tcg_pcrel_diff(s, arg);
> > + if (offset == sextract32(offset, 1, 20) << 1) {
>
> sextract64.
>
> > + /* short jump: -2097150 to 2097152 */
> > + tcg_out_opc_jump(s, OPC_JAL, link, offset);
> > + } else if (TCG_TARGET_REG_BITS == 32 ||
> > + offset == sextract32(offset, 1, 31) << 1) {
>
> sextract64.
>
> > + /* long jump: -2147483646 to 2147483648 */
> > + tcg_out_opc_upper(s, OPC_AUIPC, TCG_REG_TMP0, 0);
> > + tcg_out_opc_imm(s, OPC_JALR, link, TCG_REG_TMP0, 0);
> > + reloc_call(s->code_ptr - 2, arg);
> > + } else if (TCG_TARGET_REG_BITS == 64) {
> > + /* far jump: 64-bit */
> > + tcg_target_long imm = sextract32((tcg_target_long)arg, 0, 12);
> > + tcg_target_long base = (tcg_target_long)arg - imm;
> > + tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP0, base);
> > + tcg_out_opc_imm(s, OPC_JALR, link, TCG_REG_TMP0, imm);
>
>
> r~
- [Qemu-riscv] [RFC v1 10/23] riscv: tcg-target: Add the instruction emitters, (continued)
- [Qemu-riscv] [RFC v1 10/23] riscv: tcg-target: Add the instruction emitters, Alistair Francis, 2018/11/15
- [Qemu-riscv] [RFC v1 12/23] riscv: tcg-target: Add the mov and movi instruction, Alistair Francis, 2018/11/15
- [Qemu-riscv] [RFC v1 13/23] riscv: tcg-target: Add the extract instructions, Alistair Francis, 2018/11/15
- [Qemu-riscv] [RFC v1 14/23] riscv: tcg-target: Add the out load and store instructions, Alistair Francis, 2018/11/15
- [Qemu-riscv] [RFC v1 15/23] riscv: tcg-target: Add branch and jump instructions, Alistair Francis, 2018/11/15
[Qemu-riscv] [RFC v1 16/23] riscv: tcg-target: Add slowpath load and store instructions, Alistair Francis, 2018/11/15
[Qemu-riscv] [RFC v1 17/23] riscv: tcg-target: Add direct load and store instructions, Alistair Francis, 2018/11/15