Hi, Richard,
How about if I add and utilize "umul_highpart_di" to the libgcc
instead of expanding
multiplication for the high part directly, or add my own function with
with pre-shift, post-shift, and
64-bit constant and 64-bit operand as function parameters for division
for less than "-O3"?
On 14/06/12 19:46, Dinar Temirbulatov wrote:
> Hi,
> OK for trunk?
> thanks, Dinar.
>
I'm still not comfortable about the code bloat that this is likely to
incurr at -O2.
R.
> On Tue, Jun 12, 2012 at 11:00 AM, Paolo Bonzini wrote:
>> Il 12/06/2012 08:52, Dinar Temirbulatov ha s
Hi,
OK for trunk?
thanks, Dinar.
On Tue, Jun 12, 2012 at 11:00 AM, Paolo Bonzini wrote:
> Il 12/06/2012 08:52, Dinar Temirbulatov ha scritto:
>>> is safe? That is, that the underflows cannot produce a wrong result?
>
> [snip]
>
> Thanks very much!
>
> Paolo
2012-06-14 Dinar Temi
Il 08/06/2012 20:13, Dinar Temirbulatov ha scritto:
> that I have to add convert_to_mode () to DImode after
> emit_store_flag_force (), since emit_store_flag_force () returns
> "carry" in SImode and without convert_to_mode () call compiler fails
> with this error:
Yes, that makes sense. The new p
Hi, Paolo.
Here is the new version of patch. I have tested this version with gcc
testsuite only on i686 without new regressions, for now. Mips and arm
tests are in progress.
One strange thing I noticed:
>
> No need for this gen_reg_rtx, either, by passing a NULL_RTX target below.
>
>> + carry
Il 07/06/2012 12:21, Dinar Temirbulatov ha scritto:
> oh, I found typo in comment in the end of patch. fixed.
Great improvement, thanks!
Unfortunately we're not there yet, but much closer! I could understand
the new code much better so I suggest some more improvements below, both
to the comments
oh, I found typo in comment in the end of patch. fixed.
thanks, Dinar.
On Thu, Jun 7, 2012 at 2:14 PM, Dinar Temirbulatov
wrote:
> Hi,
> Here is new version of patch based up on Paolo review, again tested on
> arm-7l, mips-32r2 (74k), i686 without new regressions.
>
Hi,
Here is new version of patch based up on Paolo review, again tested on
arm-7l, mips-32r2 (74k), i686 without new regressions.
thanks, Dinar.
On Sat, May 26, 2012 at 4:45 PM, Paolo Bonzini wrote:
> Il 26/05/2012 14:35, Paolo Bonzini ha scritto:
>> /* We have to
Il 26/05/2012 14:35, Paolo Bonzini ha scritto:
>/* We have to return
>
> z2 + ((u0 + u1) >> GET_MODE_BITSIZE (word_mode)).
>
> u0 + u1 are the upper two words of the three-word
> intermediate result and they could have up to
> 2 * GET_MODE_BITSIZE
Il 25/05/2012 12:20, Dinar Temirbulatov ha scritto:
> + emit_store_flag_force (c, GT, u0, tmp, mode, 1, 1);
> + emit_store_flag_force (c1, GT, u1, tmp, mode, 1, 1);
> + result = expand_binop (mode, ior_optab, c, c1, cres, 1,
> OPTAB_LIB_WIDEN);
> + if (!result)
> + re
> diff --git a/gcc/config/arm/arm.c b/gcc/config/arm/arm.c
> index 2cecf45..9d6983b 100644
> --- a/gcc/config/arm/arm.c
> +++ b/gcc/config/arm/arm.c
> @@ -7131,6 +7131,8 @@ arm_rtx_costs_1 (rtx x, enum rtx_code outer, int*
> total, bool speed)
> *total = COSTS_N_INSNS (2);
>else if
Hi,
I have replaced "expand_mult" to "expand_widening_mult" and removed
all direct references to DImode, SImode modes in the
expand_mult_highpart_optab funtion. The attached patch was tested on
arm-7l, mips-32r2 (74k), i686 without new regressions. Richard, do you
think it is ready now?
On 05/22/12 07:05, Dinar Temirbulatov wrote:
> + if ((size - 1 > BITS_PER_WORD
> + && BITS_PER_WORD == 32 && mode == DImode)
Do note that this patch will not go in with hard-coded
SImode and DImode references.
Which, really, you do not even need.
&& GET_MODE_BITSIZE (mode) == 2*BITS_PER
Hi,
Here is the new version of the patch
I have fixed two errors in the previous version,
on mips32 the compiler could not expand division and terminated
with ICE, this change fixed the issue:
/* Extrating the higher part of the sum */
tmp = gen_highpart (SImode, tmp);
tmp
On 03/05/12 11:27, Dinar Temirbulatov wrote:
> Hi,
> Here is updated version of patch. I added comments describing the algorithm.
>
>> Hi Dinar. I'm afraid it gives the wrong results for some dividends
>> * 82625484914982912 / 2023346444509052928: gives 4096, should be zero
>> * 183174636040612
Hi,
Here is updated version of patch. I added comments describing the algorithm.
> Hi Dinar. I'm afraid it gives the wrong results for some dividends
> * 82625484914982912 / 2023346444509052928: gives 4096, should be zero
> * 18317463604061229328 / 2023346444509052928: gives 4109, should be 9
>
On 21 April 2012 00:57, Dinar Temirbulatov wrote:
> Hi,
> Here is the patch that adds support for divide 64-bit by constant for
> 32-bit target machines, this patch was tested on arm-7a with no new
> regressions, also I am not sure on how to avoid for example i686
> targets since div operation the
On 04/20/2012 01:57 PM, Dinar Temirbulatov wrote:
> Here is the patch that adds support for divide 64-bit by constant for
> 32-bit target machines, this patch was tested on arm-7a with no new
> regressions, also I am not sure on how to avoid for example i686
> targets since div operation there is f
18 matches
Mail list logo
