From: Vineet Gupta <vineetg@rivosinc.com>
To: Richard Biener <richard.guenther@gmail.com>
Cc: gcc-patches@gcc.gnu.org, kito.cheng@gmail.com,
Jeff Law <jeffreyalaw@gmail.com>,
Palmer Dabbelt <palmer@rivosinc.com>,
Philipp Tomsich <philipp.tomsich@vrull.eu>,
Christoph Mullner <christoph.muellner@vrull.eu>,
gnu-toolchain@rivosinc.com
Subject: [committed] expansion: make layout of x_shift*cost[][][] more efficient
Date: Fri, 21 Apr 2023 09:15:19 -0700 [thread overview]
Message-ID: <005edaa3-840f-fedb-60f1-1c19b9cd3078@rivosinc.com> (raw)
In-Reply-To: <CAFiYyc1j-0Dwy0UW3pfiL+4SP=GCUEGaL88LZCrG6JWh2i9cdw@mail.gmail.com>
On 4/19/23 00:05, Richard Biener wrote:
> On Tue, Apr 18, 2023 at 10:51 PM Vineet Gupta <vineetg@rivosinc.com> wrote:
>> when debugging expmed.[ch] for PR/108987 saw that some of the cost arrays have
>> less than ideal layout as follows:
>>
>> x_shift*cost[0..63][speed][modes]
>>
>> We would want speed to be first index since a typical compile will have
>> that fixed, followed by mode and then the shift values.
>>
>> It should be non-functional from compiler semantics pov, except
>> executing slightly faster due to better locality of shift values for
>> given speed and mode. And also a bit more intutive when debugging.
> OK, but please wait 24h in case somebody else wants to comment.
Pushed.
Thx,
-Vineet
>
> Thanks,
> Richard.
>
>> gcc/Changelog:
>>
>> * expmed.h (x_shift*_cost): convert to int [speed][mode][shift].
>> (shift*_cost_ptr ()): Access x_shift*_cost array directly.
>>
>> Signed-off-by: Vineet Gupta <vineetg@rivosinc.com>
>> ---
>> Changes since v1:
>> - Post a non stale version of patch
>> ---
>> gcc/expmed.h | 27 +++++++++++++--------------
>> 1 file changed, 13 insertions(+), 14 deletions(-)
>>
>> diff --git a/gcc/expmed.h b/gcc/expmed.h
>> index c747a0da1637..22ae1d2d0743 100644
>> --- a/gcc/expmed.h
>> +++ b/gcc/expmed.h
>> @@ -161,15 +161,14 @@ struct target_expmed {
>> struct expmed_op_cheap x_sdiv_pow2_cheap;
>> struct expmed_op_cheap x_smod_pow2_cheap;
>>
>> - /* Cost of various pieces of RTL. Note that some of these are indexed by
>> - shift count and some by mode. */
>> + /* Cost of various pieces of RTL. */
>> int x_zero_cost[2];
>> struct expmed_op_costs x_add_cost;
>> struct expmed_op_costs x_neg_cost;
>> - struct expmed_op_costs x_shift_cost[MAX_BITS_PER_WORD];
>> - struct expmed_op_costs x_shiftadd_cost[MAX_BITS_PER_WORD];
>> - struct expmed_op_costs x_shiftsub0_cost[MAX_BITS_PER_WORD];
>> - struct expmed_op_costs x_shiftsub1_cost[MAX_BITS_PER_WORD];
>> + int x_shift_cost[2][NUM_MODE_IPV_INT][MAX_BITS_PER_WORD];
>> + int x_shiftadd_cost[2][NUM_MODE_IPV_INT][MAX_BITS_PER_WORD];
>> + int x_shiftsub0_cost[2][NUM_MODE_IPV_INT][MAX_BITS_PER_WORD];
>> + int x_shiftsub1_cost[2][NUM_MODE_IPV_INT][MAX_BITS_PER_WORD];
>> struct expmed_op_costs x_mul_cost;
>> struct expmed_op_costs x_sdiv_cost;
>> struct expmed_op_costs x_udiv_cost;
>> @@ -395,8 +394,8 @@ neg_cost (bool speed, machine_mode mode)
>> inline int *
>> shift_cost_ptr (bool speed, machine_mode mode, int bits)
>> {
>> - return expmed_op_cost_ptr (&this_target_expmed->x_shift_cost[bits],
>> - speed, mode);
>> + int midx = expmed_mode_index (mode);
>> + return &this_target_expmed->x_shift_cost[speed][midx][bits];
>> }
>>
>> /* Set the COST of doing a shift in MODE by BITS when optimizing for SPEED. */
>> @@ -421,8 +420,8 @@ shift_cost (bool speed, machine_mode mode, int bits)
>> inline int *
>> shiftadd_cost_ptr (bool speed, machine_mode mode, int bits)
>> {
>> - return expmed_op_cost_ptr (&this_target_expmed->x_shiftadd_cost[bits],
>> - speed, mode);
>> + int midx = expmed_mode_index (mode);
>> + return &this_target_expmed->x_shiftadd_cost[speed][midx][bits];
>> }
>>
>> /* Set the COST of doing a shift in MODE by BITS followed by an add when
>> @@ -448,8 +447,8 @@ shiftadd_cost (bool speed, machine_mode mode, int bits)
>> inline int *
>> shiftsub0_cost_ptr (bool speed, machine_mode mode, int bits)
>> {
>> - return expmed_op_cost_ptr (&this_target_expmed->x_shiftsub0_cost[bits],
>> - speed, mode);
>> + int midx = expmed_mode_index (mode);
>> + return &this_target_expmed->x_shiftsub0_cost[speed][midx][bits];
>> }
>>
>> /* Set the COST of doing a shift in MODE by BITS and then subtracting a
>> @@ -475,8 +474,8 @@ shiftsub0_cost (bool speed, machine_mode mode, int bits)
>> inline int *
>> shiftsub1_cost_ptr (bool speed, machine_mode mode, int bits)
>> {
>> - return expmed_op_cost_ptr (&this_target_expmed->x_shiftsub1_cost[bits],
>> - speed, mode);
>> + int midx = expmed_mode_index (mode);
>> + return &this_target_expmed->x_shiftsub1_cost[speed][midx][bits];
>> }
>>
>> /* Set the COST of subtracting a shift in MODE by BITS from a value when
>> --
>> 2.34.1
>>
prev parent reply other threads:[~2023-04-21 16:15 UTC|newest]
Thread overview: 4+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-04-18 18:17 [PATCH] " Vineet Gupta
2023-04-18 20:48 ` [PATCH v2] " Vineet Gupta
2023-04-19 7:05 ` Richard Biener
2023-04-21 16:15 ` Vineet Gupta [this message]
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=005edaa3-840f-fedb-60f1-1c19b9cd3078@rivosinc.com \
--to=vineetg@rivosinc.com \
--cc=christoph.muellner@vrull.eu \
--cc=gcc-patches@gcc.gnu.org \
--cc=gnu-toolchain@rivosinc.com \
--cc=jeffreyalaw@gmail.com \
--cc=kito.cheng@gmail.com \
--cc=palmer@rivosinc.com \
--cc=philipp.tomsich@vrull.eu \
--cc=richard.guenther@gmail.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).