On Mon, Sep 19, 2011 at 09:24:47PM +0100, Stuart Brady wrote:
> On Sat, Sep 17, 2011 at 10:00:31PM +0200, Stefan Weil wrote:
>
[...]
> > +u64 = ((helper_function)t0)(tci_read_reg(TCG_REG_R0),
> > +tci_read_reg(TCG_REG_R1),
> > +
On Sat, Sep 17, 2011 at 10:00:31PM +0200, Stefan Weil wrote:
> +#if MAX_OPC_PARAM_IARGS != 4
> +# error Fix needed, number of supported input arguments changed!
> +#endif
> +#if TCG_TARGET_REG_BITS == 32
> +typedef uint64_t (*helper_function)(tcg_target_ulong, tcg_target_ulong,
> +
On 09/17/2011 01:00 PM, Stefan Weil wrote:
> +#if TCG_TARGET_HAS_ext8s_i32
> +case INDEX_op_ext8s_i32:
> +t0 = *tb_ptr++;
> +t1 = tci_read_r8s(&tb_ptr);
> +tci_write_reg32(t0, t1);
> +break;
> +#endif
You really ought not need all these ifdef
> You generally want CSE, yes? So you can't blame gcc for getting it
> wrong sometimes.
There are cases where CSE pessimizes the code, .e.g when it increases
memory pressure too much or caches something that is easier recomputed.
This is just another one.
BTW I checked again and the problem see
On 09/19/2011 09:52 AM, Andi Kleen wrote:
> I think it also improves branch target prediction - if you have a tight
> loop of a few opcodes the predictor can guess where you're headed (since
> there is a separate lookup key for each opcode), whereas with the
> original code, there's a single
> I think it also improves branch target prediction - if you have a tight
> loop of a few opcodes the predictor can guess where you're headed (since
> there is a separate lookup key for each opcode), whereas with the
> original code, there's a single key which cannot be used to predict the
> branc
On 09/18/2011 10:22 AM, Paolo Bonzini wrote:
On 09/18/2011 07:49 AM, Stefan Weil wrote:
Is there really any difference in the generated code?
gcc already uses a jump table internally to handle the
switch cases.
You typically save something on range checks, and it enables a lot
more tricks for
On 09/18/2011 07:49 AM, Stefan Weil wrote:
Is there really any difference in the generated code?
gcc already uses a jump table internally to handle the
switch cases.
You typically save something on range checks, and it enables a lot more
tricks for use later (e.g. using multiple jump tables to
Am 18.09.2011 06:03, schrieb Andi Kleen:
Stefan Weil writes:
+
+ switch (opc) {
+ case INDEX_op_end:
+ case INDEX_op_nop:
+ break;
You could probably get some more speed out of this by using a threaded
interpreter with gcc's computed goto extension. That's typically
significantly faster than
Stefan Weil writes:
> +
> +switch (opc) {
> +case INDEX_op_end:
> +case INDEX_op_nop:
> +break;
You could probably get some more speed out of this by using a threaded
interpreter with gcc's computed goto extension. That's typically
significantly faster than a p
Signed-off-by: Stefan Weil
---
tcg/tcg.h |4 +-
tcg/tci.c | 1200 +
2 files changed, 1203 insertions(+), 1 deletions(-)
create mode 100644 tcg/tci.c
diff --git a/tcg/tcg.h b/tcg/tcg.h
index 1859fae..c99c7ea 100644
--- a/tcg/tcg.h
+
11 matches
Mail list logo
