From: Alexander Monakov <amonakov@ispras.ru>
To: gcc-patches@gcc.gnu.org
Cc: "Jan Hubička" <honza.hubicka@gmail.com>,
"Joshi, Tejas Sanjay" <TejasSanjay.Joshi@amd.com>,
"Kumar, Venkataramanan" <Venkataramanan.Kumar@amd.com>,
"Alexander Monakov" <amonakov@ispras.ru>
Subject: [PATCH 1/2] i386: correct x87&SSE division modeling in znver.md
Date: Tue, 1 Nov 2022 19:26:36 +0300 [thread overview]
Message-ID: <20221101162637.14238-2-amonakov@ispras.ru> (raw)
In-Reply-To: <20221101162637.14238-1-amonakov@ispras.ru>
Correct modeling of division instructions in the SIMD/FP domain for
AMD Zen architectures and avoid combinatorial explosion of automaton
tables by modeling the separate floating-point division unit and
correcting reservations to reflect reciprocal throughput of the
corresponding instructions, similar to earlier commit
5cee5f94000 ("i386: correct integer division modeling in znver.md").
Division is partially pipelined and some instructions have fractional
throughput (e.g. Zen 3 can issue divss and divsd each 3.5 and 4.5
cycles on average, respectively). Considering these CPUs implement
out-of-order execution, the model doesn't need to be exact to the last
cycle, so simplify it by using 4/5 cycles for SF/DF modes, and not
modeling the fact that FP3 pipe is occupied for one cycle.
Top znver table sizes in insn-automata.o:
Before:
428108 r znver1_fp_min_issue_delay
856216 r znver1_fp_transitions
After:
30056 r znver1_fp_min_issue_delay
120224 r znver1_fp_transitions
gcc/ChangeLog:
PR target/87832
* config/i386/znver.md (znver1_fdiv): New automaton.
(znver1-fdiv): New unit.
(znver1_fp_op_div): Correct unit and cycles in the reservation.
(znver1_fp_op_div_load): Ditto.
(znver1_fp_op_idiv_load): Ditto.
(znver2_fp_op_idiv_load): Ditto.
(znver1_ssediv_ss_ps): Ditto.
(znver1_ssediv_ss_ps_load): Ditto.
(znver1_ssediv_sd_pd): Ditto.
(znver1_ssediv_sd_pd_load): Ditto.
(znver1_ssediv_avx256_ps): Ditto.
(znver1_ssediv_avx256_ps_load): Ditto.
(znver1_ssediv_avx256_pd): Ditto.
(znver1_ssediv_avx256_pd_load): Ditto.
---
gcc/config/i386/znver.md | 27 ++++++++++++++-------------
1 file changed, 14 insertions(+), 13 deletions(-)
diff --git a/gcc/config/i386/znver.md b/gcc/config/i386/znver.md
index 4aa098fd8..c52f8b532 100644
--- a/gcc/config/i386/znver.md
+++ b/gcc/config/i386/znver.md
@@ -24,7 +24,7 @@ (define_attr "znver1_decode" "direct,vector,double"
;; AMD znver1, znver2 and znver3 Scheduling
;; Modeling automatons for zen decoders, integer execution pipes,
;; SIMD/FP domain, AGU pipes, and dividers.
-(define_automaton "znver1, znver1_ieu, znver1_fp, znver1_agu, znver1_idiv")
+(define_automaton "znver1, znver1_ieu, znver1_fp, znver1_agu, znver1_idiv, znver1_fdiv")
;; Decoders unit has 4 decoders and all of them can decode fast path
;; and vector type instructions.
@@ -95,6 +95,7 @@ (define_reservation "znver2-fvector" "znver1-fp0+znver1-fp1
;; Dividers
(define_cpu_unit "znver1-idiv" "znver1_idiv")
+(define_cpu_unit "znver1-fdiv" "znver1_fdiv")
;; Call instruction
(define_insn_reservation "znver1_call" 1
@@ -591,27 +592,27 @@ (define_insn_reservation "znver1_fp_op_div" 15
(and (eq_attr "cpu" "znver1,znver2,znver3")
(and (eq_attr "type" "fdiv")
(eq_attr "memory" "none")))
- "znver1-direct,znver1-fp3*15")
+ "znver1-direct,znver1-fdiv*6")
(define_insn_reservation "znver1_fp_op_div_load" 22
(and (eq_attr "cpu" "znver1,znver2,znver3")
(and (eq_attr "type" "fdiv")
(eq_attr "memory" "load")))
- "znver1-direct,znver1-load,znver1-fp3*15")
+ "znver1-direct,znver1-load,znver1-fdiv*6")
(define_insn_reservation "znver1_fp_op_idiv_load" 27
(and (eq_attr "cpu" "znver1")
(and (eq_attr "type" "fdiv")
(and (eq_attr "fp_int_src" "true")
(eq_attr "memory" "load"))))
- "znver1-double,znver1-load,znver1-fp3*19")
+ "znver1-double,znver1-load,znver1-fdiv*6")
(define_insn_reservation "znver2_fp_op_idiv_load" 26
(and (eq_attr "cpu" "znver2,znver3")
(and (eq_attr "type" "fdiv")
(and (eq_attr "fp_int_src" "true")
(eq_attr "memory" "load"))))
- "znver1-double,znver1-load,znver1-fp3*19")
+ "znver1-double,znver1-load,znver1-fdiv*6")
;; MMX, SSE, SSEn.n, AVX, AVX2 instructions
@@ -1088,7 +1089,7 @@ (define_insn_reservation "znver1_ssediv_ss_ps" 10
(eq_attr "mode" "V8SF,V4SF,SF")))
(and (eq_attr "type" "ssediv")
(eq_attr "memory" "none")))
- "znver1-direct,znver1-fp3*10")
+ "znver1-direct,znver1-fdiv*4")
(define_insn_reservation "znver1_ssediv_ss_ps_load" 17
(and (ior (and (eq_attr "cpu" "znver1")
@@ -1099,7 +1100,7 @@ (define_insn_reservation "znver1_ssediv_ss_ps_load" 17
(eq_attr "mode" "V8SF,V4SF,SF")))
(and (eq_attr "type" "ssediv")
(eq_attr "memory" "load")))
- "znver1-direct,znver1-load,znver1-fp3*10")
+ "znver1-direct,znver1-load,znver1-fdiv*4")
(define_insn_reservation "znver1_ssediv_sd_pd" 13
(and (ior (and (eq_attr "cpu" "znver1")
@@ -1110,7 +1111,7 @@ (define_insn_reservation "znver1_ssediv_sd_pd" 13
(eq_attr "mode" "V4DF,V2DF,DF")))
(and (eq_attr "type" "ssediv")
(eq_attr "memory" "none")))
- "znver1-direct,znver1-fp3*13")
+ "znver1-direct,znver1-fdiv*5")
(define_insn_reservation "znver1_ssediv_sd_pd_load" 20
(and (ior (and (eq_attr "cpu" "znver1")
@@ -1121,35 +1122,35 @@ (define_insn_reservation "znver1_ssediv_sd_pd_load" 20
(eq_attr "mode" "V4DF,V2DF,DF")))
(and (eq_attr "type" "ssediv")
(eq_attr "memory" "load")))
- "znver1-direct,znver1-load,znver1-fp3*13")
+ "znver1-direct,znver1-load,znver1-fdiv*5")
(define_insn_reservation "znver1_ssediv_avx256_ps" 12
(and (eq_attr "cpu" "znver1")
(and (eq_attr "mode" "V8SF")
(and (eq_attr "memory" "none")
(eq_attr "type" "ssediv"))))
- "znver1-double,znver1-fp3*12")
+ "znver1-double,znver1-fdiv*8")
(define_insn_reservation "znver1_ssediv_avx256_ps_load" 19
(and (eq_attr "cpu" "znver1")
(and (eq_attr "mode" "V8SF")
(and (eq_attr "type" "ssediv")
(eq_attr "memory" "load"))))
- "znver1-double,znver1-load,znver1-fp3*12")
+ "znver1-double,znver1-load,znver1-fdiv*8")
(define_insn_reservation "znver1_ssediv_avx256_pd" 15
(and (eq_attr "cpu" "znver1")
(and (eq_attr "mode" "V4DF")
(and (eq_attr "type" "ssediv")
(eq_attr "memory" "none"))))
- "znver1-double,znver1-fp3*15")
+ "znver1-double,znver1-fdiv*10")
(define_insn_reservation "znver1_ssediv_avx256_pd_load" 22
(and (eq_attr "cpu" "znver1")
(and (eq_attr "mode" "V4DF")
(and (eq_attr "type" "ssediv")
(eq_attr "memory" "load"))))
- "znver1-double,znver1-load,znver1-fp3*15")
+ "znver1-double,znver1-load,znver1-fdiv*10")
;; SSE MUL
(define_insn_reservation "znver1_ssemul_ss_ps" 3
(and (ior (and (eq_attr "cpu" "znver1")
--
2.37.2
next prev parent reply other threads:[~2022-11-01 16:26 UTC|newest]
Thread overview: 9+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-11-01 16:26 [PATCH 0/2] i386: slim down insn-automata [PR 87832] Alexander Monakov
2022-11-01 16:26 ` Alexander Monakov [this message]
2022-11-01 16:26 ` [PATCH 2/2] i386: correct x87&SSE multiplication modeling in znver.md Alexander Monakov
2022-11-16 11:53 ` Kumar, Venkataramanan
2022-11-16 12:21 ` Jan Hubička
2022-11-16 13:13 ` Alexander Monakov
2022-11-16 13:28 ` Jan Hubička
2022-11-07 11:27 ` [PATCH 0/2] i386: slim down insn-automata [PR 87832] Alexander Monakov
2022-11-14 11:19 ` Alexander Monakov
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=20221101162637.14238-2-amonakov@ispras.ru \
--to=amonakov@ispras.ru \
--cc=TejasSanjay.Joshi@amd.com \
--cc=Venkataramanan.Kumar@amd.com \
--cc=gcc-patches@gcc.gnu.org \
--cc=honza.hubicka@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).