From: Jiufu Guo <guojiufu@linux.ibm.com>
To: Segher Boessenkool <segher@kernel.crashing.org>
Cc: gcc-patches@gcc.gnu.org, dje.gcc@gmail.com, linkw@gcc.gnu.org
Subject: Re: [PATCH V6] rs6000: Optimize cmp on rotated 16bits constant
Date: Wed, 14 Dec 2022 16:26:54 +0800 [thread overview]
Message-ID: <7e8rjajsq9.fsf@pike.rch.stglabs.ibm.com> (raw)
In-Reply-To: <20221213210232.GM25951@gate.crashing.org> (Segher Boessenkool's message of "Tue, 13 Dec 2022 15:02:32 -0600")
Hi,
Segher Boessenkool <segher@kernel.crashing.org> writes:
> Hi!
>
> Sorry for the tardiness.
>
> On Mon, Aug 29, 2022 at 11:42:16AM +0800, Jiufu Guo wrote:
>> When checking eq/ne with a constant which has only 16bits, it can be
>> optimized to check the rotated data. By this, the constant building
>> is optimized.
>>
>> As the example in PR103743:
>> For "in == 0x8000000000000000LL", this patch generates:
>> rotldi %r3,%r3,16
>> cmpldi %cr0,%r3,32768
>> instead:
>> li %r9,-1
>> rldicr %r9,%r9,0,0
>> cmpd %cr0,%r3,%r9
>
> FWIW, I find the winnt assembler syntax very hard to read, and I doubt
> I am the only one.
Oh, sorry about that. I will avoid to add '-mregnames' to dump asm. :)
BTW, what options are you used to dump asm code?
>
> So you're doing
> rotldi 3,3,16 ; cmpldi 3,0x8000
> instead of
> li 9,-1 ; rldicr 9,9,0,0 ; cmpd 3,9
>
>> +/* Check if C can be rotated from an immediate which starts (as 64bit integer)
>> + with at least CLZ bits zero.
>> +
>> + Return the number by which C can be rotated from the immediate.
>> + Return -1 if C can not be rotated as from. */
>> +
>> +int
>> +rotate_from_leading_zeros_const (unsigned HOST_WIDE_INT c, int clz)
>
> The name does not say what the function does. Can you think of a better
> name?
>
> Maybe it is better to not return magic values anyway? So perhaps
>
> bool
> can_be_done_as_compare_of_rotate (unsigned HOST_WIDE_INT c, int clz, int *rot)
>
> (with *rot written if the return value is true).
Thanks for your suggestion!
It is checking if a constant can be rotated from/to a value which has
only few tailing nonzero bits (all leading bits are zeros).
So, I'm thinking to name the function as something like:
can_be_rotated_to_lowbits.
>
>> + /* case c. xx10.....0xx: rotate 'clz + 1' bits firstly, then check case b.
>
> s/firstly/first/
Thanks!
>
>> +/* Check if C can be rotated from an immediate operand of cmpdi or cmpldi. */
>> +
>> +bool
>> +compare_rotate_immediate_p (unsigned HOST_WIDE_INT c)
>
> No _p please, this function is not a predicate (at least, the name does
> not say what it tests). So a better name please. This matters even
> more for extern functions (like this one) because the function
> implementation is always farther away so you do not easily have all
> interface details in mind. Good names help :-)
Thanks! Name is always a matter. :)
Maybe we can name this funciton as "can_be_rotated_as_compare_operand",
or "is_constant_rotateable_for_compare", because this function checks
"if a constant can be rotated to/from an immediate operand of
cmpdi/cmpldi".
>
>> +(define_code_iterator eqne [eq ne])
>> +(define_code_attr EQNE [(eq "EQ") (ne "NE")])
>
> Just <CODE> or <CODE:eqne> should work?
Great! Thanks for point out this! <eqne:CODE> works.
>
> Please fix these things. Almost there :-)
I updated the patch as below. Bootstraping and regtesting is ongoing.
Thanks again for your careful and insight review!
BR,
Jeff (Jiufu)
----------
When checking eq/ne with a constant which has only 16bits, it can be
optimized to check the rotated data. By this, the constant building
is optimized.
As the example in PR103743:
For "in == 0x8000000000000000LL", this patch generates:
rotldi 3,3,1 ; cmpldi 0,3,1
instead of:
li 9,-1 ; rldicr 9,9,0,0 ; cmpd 0,3,9
Compare with previous version:
This patch refactor the code according to review comments.
e.g. updating function names/comments/code.
PR target/103743
gcc/ChangeLog:
* config/rs6000/rs6000-protos.h (can_be_rotated_to_lowbits): New.
(can_be_rotated_as_compare_operand): New.
* config/rs6000/rs6000.cc (can_be_rotated_to_lowbits): New definition.
(can_be_rotated_as_compare_operand): New definition.
* config/rs6000/rs6000.md (*rotate_on_cmpdi): New define_insn_and_split.
gcc/testsuite/ChangeLog:
* gcc.target/powerpc/pr103743.c: New test.
* gcc.target/powerpc/pr103743_1.c: New test.
---
gcc/config/rs6000/rs6000-protos.h | 2 +
gcc/config/rs6000/rs6000.cc | 56 +++++++++++
gcc/config/rs6000/rs6000.md | 63 +++++++++++-
gcc/testsuite/gcc.target/powerpc/pr103743.c | 52 ++++++++++
gcc/testsuite/gcc.target/powerpc/pr103743_1.c | 95 +++++++++++++++++++
5 files changed, 267 insertions(+), 1 deletion(-)
create mode 100644 gcc/testsuite/gcc.target/powerpc/pr103743.c
create mode 100644 gcc/testsuite/gcc.target/powerpc/pr103743_1.c
diff --git a/gcc/config/rs6000/rs6000-protos.h b/gcc/config/rs6000/rs6000-protos.h
index d0d89320ef6..9626917e359 100644
--- a/gcc/config/rs6000/rs6000-protos.h
+++ b/gcc/config/rs6000/rs6000-protos.h
@@ -35,6 +35,8 @@ extern bool xxspltib_constant_p (rtx, machine_mode, int *, int *);
extern int vspltis_shifted (rtx);
extern HOST_WIDE_INT const_vector_elt_as_int (rtx, unsigned int);
extern bool macho_lo_sum_memory_operand (rtx, machine_mode);
+extern bool can_be_rotated_to_lowbits (unsigned HOST_WIDE_INT, int, int *);
+extern bool can_be_rotated_as_compare_operand (unsigned HOST_WIDE_INT);
extern int num_insns_constant (rtx, machine_mode);
extern int small_data_operand (rtx, machine_mode);
extern bool mem_operand_gpr (rtx, machine_mode);
diff --git a/gcc/config/rs6000/rs6000.cc b/gcc/config/rs6000/rs6000.cc
index b3a609f3aa3..2b0df8479f0 100644
--- a/gcc/config/rs6000/rs6000.cc
+++ b/gcc/config/rs6000/rs6000.cc
@@ -14925,6 +14925,62 @@ rs6000_reverse_condition (machine_mode mode, enum rtx_code code)
return reverse_condition (code);
}
+/* Check if C (as 64bit integer) can be rotated to a constant which constains
+ nonzero bits at LOWBITS only.
+
+ Return true if C can be rotated to such constant. And *ROT is written to
+ the number by which C is rotated.
+ Return false otherwise. */
+
+bool
+can_be_rotated_to_lowbits (unsigned HOST_WIDE_INT c, int lowbits, int *rot)
+{
+ int clz = HOST_BITS_PER_WIDE_INT - lowbits;
+
+ /* case a. 0..0xxx: already at least clz zeros. */
+ int lz = clz_hwi (c);
+ if (lz >= clz)
+ {
+ *rot = 0;
+ return true;
+ }
+
+ /* case b. 0..0xxx0..0: at least clz zeros. */
+ int tz = ctz_hwi (c);
+ if (lz + tz >= clz)
+ {
+ *rot = HOST_BITS_PER_WIDE_INT - tz;
+ return true;
+ }
+
+ /* case c. xx10.....0xx: rotate 'clz - 1' bits first, then check case b.
+ ^bit -> Vbit, , then zeros are at head or tail.
+ 00...00xxx100, 'clz - 1' >= 'bits of xxxx'. */
+ const int rot_bits = lowbits + 1;
+ unsigned HOST_WIDE_INT rc = (c >> rot_bits) | (c << (clz - 1));
+ tz = ctz_hwi (rc);
+ if (clz_hwi (rc) + tz >= clz)
+ {
+ *rot = HOST_BITS_PER_WIDE_INT - (tz + rot_bits);
+ return true;
+ }
+
+ return false;
+}
+
+/* Check if C can be rotated to an immediate operand of cmpdi or cmpldi. */
+
+bool
+can_be_rotated_as_compare_operand (unsigned HOST_WIDE_INT c)
+{
+ int rot = 0;
+ /* leading 48 zeros + 16 lowbits (cmpldi),
+ or leading 49 ones + 15 lowbits (cmpdi). */
+ bool res = can_be_rotated_to_lowbits (~c, 15, &rot)
+ || can_be_rotated_to_lowbits (c, 16, &rot);
+ return res && rot > 0;
+}
+
/* Generate a compare for CODE. Return a brand-new rtx that
represents the result of the compare. */
diff --git a/gcc/config/rs6000/rs6000.md b/gcc/config/rs6000/rs6000.md
index 4bd1dfd3da9..dabfe5dfdfd 100644
--- a/gcc/config/rs6000/rs6000.md
+++ b/gcc/config/rs6000/rs6000.md
@@ -7765,6 +7765,68 @@ (define_insn "*movsi_from_df"
"xscvdpsp %x0,%x1"
[(set_attr "type" "fp")])
+
+(define_code_iterator eqne [eq ne])
+
+;; "i == C" ==> "rotl(i,N) == rotl(C,N)"
+(define_insn_and_split "*rotate_on_cmpdi"
+ [(set (pc)
+ (if_then_else (eqne (match_operand:DI 1 "gpc_reg_operand" "r")
+ (match_operand:DI 2 "const_int_operand" "n"))
+ (label_ref (match_operand 0 ""))
+ (pc)))
+ (clobber (match_scratch:DI 3 "=r"))
+ (clobber (match_scratch:CCUNS 4 "=y"))]
+ "TARGET_POWERPC64 && num_insns_constant (operands[2], DImode) > 1
+ && can_be_rotated_as_compare_operand (UINTVAL (operands[2]))"
+ "#"
+ "&& 1"
+ [(pc)]
+{
+ rtx note = find_reg_note (curr_insn, REG_BR_PROB, 0);
+ bool sgn = false;
+ unsigned HOST_WIDE_INT C = INTVAL (operands[2]);
+ int rot;
+ if (!can_be_rotated_to_lowbits (C, 16, &rot))
+ {
+ /* cmpdi */
+ sgn = true;
+ bool res = can_be_rotated_to_lowbits (~C, 15, &rot);
+ gcc_assert (res);
+ }
+ rtx n = GEN_INT (rot);
+
+ /* i' = rotl (i, n) */
+ rtx op0 = can_create_pseudo_p () ? gen_reg_rtx (DImode) : operands[3];
+ emit_insn (gen_rtx_SET (op0, gen_rtx_ROTATE (DImode, operands[1], n)));
+
+ /* C' = rotl (C, n) */
+ rtx op1 = GEN_INT ((C << rot) | (C >> (HOST_BITS_PER_WIDE_INT - rot)));
+
+ /* i' == C' */
+ machine_mode comp_mode = sgn ? CCmode : CCUNSmode;
+ rtx cc = can_create_pseudo_p () ? gen_reg_rtx (comp_mode) : operands[4];
+ PUT_MODE (cc, comp_mode);
+ emit_insn (gen_rtx_SET (cc, gen_rtx_COMPARE (comp_mode, op0, op1)));
+ rtx cmp = gen_rtx_<eqne:CODE> (CCmode, cc, const0_rtx);
+ rtx loc_ref = gen_rtx_LABEL_REF (VOIDmode, operands[0]);
+ emit_jump_insn (gen_rtx_SET (pc_rtx,
+ gen_rtx_IF_THEN_ELSE (VOIDmode, cmp,
+ loc_ref, pc_rtx)));
+
+ /* keep the probability info for the prediction of the branch insn. */
+ if (note)
+ {
+ profile_probability prob
+ = profile_probability::from_reg_br_prob_note (XINT (note, 0));
+
+ add_reg_br_prob_note (get_last_insn (), prob);
+ }
+
+ DONE;
+}
+)
+
;; Split a load of a large constant into the appropriate two-insn
;; sequence.
@@ -13453,7 +13515,6 @@ (define_expand "@ctr<mode>"
;; rs6000_legitimate_combined_insn prevents combine creating any of
;; the ctr<mode> insns.
-(define_code_iterator eqne [eq ne])
(define_code_attr bd [(eq "bdz") (ne "bdnz")])
(define_code_attr bd_neg [(eq "bdnz") (ne "bdz")])
diff --git a/gcc/testsuite/gcc.target/powerpc/pr103743.c b/gcc/testsuite/gcc.target/powerpc/pr103743.c
new file mode 100644
index 00000000000..41c686bb4cb
--- /dev/null
+++ b/gcc/testsuite/gcc.target/powerpc/pr103743.c
@@ -0,0 +1,52 @@
+/* { dg-options "-O2" } */
+/* { dg-do compile { target has_arch_ppc64 } } */
+
+/* { dg-final { scan-assembler-times {\mcmpldi\M} 10 } } */
+/* { dg-final { scan-assembler-times {\mcmpdi\M} 4 } } */
+/* { dg-final { scan-assembler-times {\mrotldi\M} 14 } } */
+
+int foo (int a);
+
+int __attribute__ ((noinline)) udi_fun (unsigned long long in)
+{
+ if (in == (0x8642000000000000ULL))
+ return foo (1);
+ if (in == (0x7642000000000000ULL))
+ return foo (12);
+ if (in == (0x8000000000000000ULL))
+ return foo (32);
+ if (in == (0x8700000000000091ULL))
+ return foo (33);
+ if (in == (0x8642FFFFFFFFFFFFULL))
+ return foo (46);
+ if (in == (0x7642FFFFFFFFFFFFULL))
+ return foo (51);
+ if (in == (0x7567000000ULL))
+ return foo (9);
+ if (in == (0xFFF8567FFFFFFFFFULL))
+ return foo (19);
+
+ return 0;
+}
+
+int __attribute__ ((noinline)) di_fun (long long in)
+{
+ if (in == (0x8642000000000000LL))
+ return foo (1);
+ if (in == (0x7642000000000000LL))
+ return foo (12);
+ if (in == (0x8000000000000000LL))
+ return foo (32);
+ if (in == (0x8700000000000091LL))
+ return foo (33);
+ if (in == (0x8642FFFFFFFFFFFFLL))
+ return foo (46);
+ if (in == (0x7642FFFFFFFFFFFFLL))
+ return foo (51);
+ if (in == (0x7567000000LL))
+ return foo (9);
+ if (in == (0xFFF8567FFFFFFFFFLL))
+ return foo (19);
+
+ return 0;
+}
diff --git a/gcc/testsuite/gcc.target/powerpc/pr103743_1.c b/gcc/testsuite/gcc.target/powerpc/pr103743_1.c
new file mode 100644
index 00000000000..e128aae7574
--- /dev/null
+++ b/gcc/testsuite/gcc.target/powerpc/pr103743_1.c
@@ -0,0 +1,95 @@
+/* { dg-do run } */
+/* { dg-options "-O2 -std=c99" } */
+
+int
+foo (int a)
+{
+ return a + 6;
+}
+
+int __attribute__ ((noipa)) udi_fun (unsigned long long in)
+{
+ if (in == (0x8642000000000000ULL))
+ return foo (1);
+ if (in == (0x7642000000000000ULL))
+ return foo (12);
+ if (in == (0x8000000000000000ULL))
+ return foo (32);
+ if (in == (0x8700000000000091ULL))
+ return foo (33);
+ if (in == (0x8642FFFFFFFFFFFFULL))
+ return foo (46);
+ if (in == (0x7642FFFFFFFFFFFFULL))
+ return foo (51);
+ if (in == (0x7567000000ULL))
+ return foo (9);
+ if (in == (0xFFF8567FFFFFFFFFULL))
+ return foo (19);
+
+ return 0;
+}
+
+int __attribute__ ((noipa)) di_fun (long long in)
+{
+ if (in == (0x8642000000000000LL))
+ return foo (1);
+ if (in == (0x7642000000000000LL))
+ return foo (12);
+ if (in == (0x8000000000000000LL))
+ return foo (32);
+ if (in == (0x8700000000000091LL))
+ return foo (33);
+ if (in == (0x8642FFFFFFFFFFFFLL))
+ return foo (46);
+ if (in == (0x7642FFFFFFFFFFFFLL))
+ return foo (51);
+ return 0;
+}
+
+int
+main ()
+{
+ int e = 0;
+ if (udi_fun (6) != 0)
+ e++;
+ if (udi_fun (0x8642000000000000ULL) != foo (1))
+ e++;
+ if (udi_fun (0x7642000000000000ULL) != foo (12))
+ e++;
+ if (udi_fun (0x8000000000000000ULL) != foo (32))
+ e++;
+ if (udi_fun (0x8700000000000091ULL) != foo (33))
+ e++;
+ if (udi_fun (0x8642FFFFFFFFFFFFULL) != foo (46))
+ e++;
+ if (udi_fun (0x7642FFFFFFFFFFFFULL) != foo (51))
+ e++;
+ if (udi_fun (0x7567000000ULL) != foo (9))
+ e++;
+ if (udi_fun (0xFFF8567FFFFFFFFFULL) != foo (19))
+ e++;
+
+ if (di_fun (6) != 0)
+ e++;
+ if (di_fun (0x8642000000000000LL) != foo (1))
+ e++;
+ if (di_fun (0x7642000000000000LL) != foo (12))
+ e++;
+ if (di_fun (0x8000000000000000LL) != foo (32))
+ e++;
+ if (di_fun (0x8700000000000091LL) != foo (33))
+ e++;
+ if (di_fun (0x8642FFFFFFFFFFFFLL) != foo (46))
+ e++;
+ if (di_fun (0x7642FFFFFFFFFFFFLL) != foo (51))
+ e++;
+ if (udi_fun (0x7567000000LL) != foo (9))
+ e++;
+ if (udi_fun (0xFFF8567FFFFFFFFFLL) != foo (19))
+ e++;
+
+ if (e)
+ __builtin_abort ();
+ return 0;
+}
+
--
2.17.1
>
>
> Segher
next prev parent reply other threads:[~2022-12-14 8:27 UTC|newest]
Thread overview: 11+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-08-29 3:42 Jiufu Guo
2022-09-15 7:35 ` Ping: " Jiufu Guo
2022-10-12 6:42 ` Ping^2: " Jiufu Guo
2022-11-09 3:01 ` Ping^3: " Jiufu Guo
2022-12-02 2:45 ` Ping^4: " Jiufu Guo
2022-12-13 3:49 ` Ping^5: " Jiufu Guo
2022-12-13 21:02 ` Segher Boessenkool
2022-12-14 8:26 ` Jiufu Guo [this message]
2022-12-15 2:28 ` Jiufu Guo
2022-12-16 20:36 ` Segher Boessenkool
2022-12-19 14:10 ` Jiufu Guo
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=7e8rjajsq9.fsf@pike.rch.stglabs.ibm.com \
--to=guojiufu@linux.ibm.com \
--cc=dje.gcc@gmail.com \
--cc=gcc-patches@gcc.gnu.org \
--cc=linkw@gcc.gnu.org \
--cc=segher@kernel.crashing.org \
/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).