On Mon, 31 Oct 2022, Jan Hubička wrote:
> Hello,
> thanks for checking the performance. The patch is OK.
Thanks, pushed the attached patch, and working on a corresponding change for
floating-point divisions.
Alexander
From 1962a8b22d3d3fb5b6bb5598295a4571daf8876f Mon Sep 17 00:00:00 2001
From: Alexander Monakov <amona...@ispras.ru>
Date: Mon, 31 Oct 2022 17:35:57 +0300
Subject: [PATCH] i386: correct integer division modeling in znver.md
In znver.md, division instructions have descriptions like
(define_insn_reservation "znver1_idiv_DI" 41
(and (eq_attr "cpu" "znver1,znver2")
(and (eq_attr "type" "idiv")
(and (eq_attr "mode" "DI")
(eq_attr "memory" "none"))))
"znver1-double,znver1-ieu2*41")
which says that DImode idiv has latency 41 (which is correct) and that
it occupies 2nd integer execution unit for 41 consecutive cycles, but
that is not correct:
1) the division instruction is partially pipelined, and has throughput
1/14, not 1/41;
2) for the most part it occupies a separate division unit, not the
general arithmetic unit.
Evidently, interaction of such 41-cycle paths with the rest of
reservations causes a combinatorial explosion in the automaton.
Fix this by modeling the integer division unit properly, and correcting
reservations to use the measured reciprocal throughput of those
instructions (available from uops.info). A similar correction for
floating-point divisions is left for a followup patch.
Top 5 znver table sizes, before:
68692 r znver1_ieu_check
68692 r znver1_ieu_transitions
99792 r znver1_ieu_min_issue_delay
428108 r znver1_fp_min_issue_delay
856216 r znver1_fp_transitions
After:
1454 r znver1_ieu_translate
1454 r znver1_translate
2304 r znver1_ieu_transitions
428108 r znver1_fp_min_issue_delay
856216 r znver1_fp_transitions
gcc/ChangeLog:
PR target/87832
* config/i386/znver.md (znver1_idiv): New automaton.
(znver1-idiv): New unit.
(znver1_idiv_DI): Correct unit and cycles in the reservation.
(znver1_idiv_SI): Ditto.
(znver1_idiv_HI): Ditto.
(znver1_idiv_QI): Ditto.
(znver1_idiv_mem_DI): Ditto.
(znver1_idiv_mem_SI): Ditto.
(znver1_idiv_mem_HI): Ditto.
(znver1_idiv_mem_QI): Ditto.
(znver3_idiv_DI): Ditto.
(znver3_idiv_SI): Ditto.
(znver3_idiv_HI): Ditto.
(znver3_idiv_QI): Ditto.
(znver3_idiv_mem_DI): Ditto.
(znver3_idiv_mem_SI): Ditto.
(znver3_idiv_mem_HI): Ditto.
(znver3_idiv_mem_QI): Ditto.
---
gcc/config/i386/znver.md | 39 +++++++++++++++++++++------------------
1 file changed, 21 insertions(+), 18 deletions(-)
diff --git a/gcc/config/i386/znver.md b/gcc/config/i386/znver.md
index 9c25b4e27..4aa098fd8 100644
--- a/gcc/config/i386/znver.md
+++ b/gcc/config/i386/znver.md
@@ -23,8 +23,8 @@ (define_attr "znver1_decode" "direct,vector,double"
;; AMD znver1, znver2 and znver3 Scheduling
;; Modeling automatons for zen decoders, integer execution pipes,
-;; AGU pipes and floating point execution units.
-(define_automaton "znver1, znver1_ieu, znver1_fp, znver1_agu")
+;; SIMD/FP domain, AGU pipes, and dividers.
+(define_automaton "znver1, znver1_ieu, znver1_fp, znver1_agu, znver1_idiv")
;; Decoders unit has 4 decoders and all of them can decode fast path
;; and vector type instructions.
@@ -93,6 +93,9 @@ (define_reservation "znver2-fvector" "znver1-fp0+znver1-fp1
+znver1-fp2+znver1-fp3
+znver1-agu0+znver1-agu1+znver2-agu2")
+;; Dividers
+(define_cpu_unit "znver1-idiv" "znver1_idiv")
+
;; Call instruction
(define_insn_reservation "znver1_call" 1
(and (eq_attr "cpu" "znver1")
@@ -176,28 +179,28 @@ (define_insn_reservation "znver1_idiv_DI" 41
(and (eq_attr "type" "idiv")
(and (eq_attr "mode" "DI")
(eq_attr "memory" "none"))))
- "znver1-double,znver1-ieu2*41")
+ "znver1-double,znver1-idiv*14")
(define_insn_reservation "znver1_idiv_SI" 25
(and (eq_attr "cpu" "znver1,znver2")
(and (eq_attr "type" "idiv")
(and (eq_attr "mode" "SI")
(eq_attr "memory" "none"))))
- "znver1-double,znver1-ieu2*25")
+ "znver1-double,znver1-idiv*14")
(define_insn_reservation "znver1_idiv_HI" 17
(and (eq_attr "cpu" "znver1,znver2")
(and (eq_attr "type" "idiv")
(and (eq_attr "mode" "HI")
(eq_attr "memory" "none"))))
- "znver1-double,znver1-ieu2*17")
+ "znver1-double,znver1-idiv*14")
(define_insn_reservation "znver1_idiv_QI" 12
(and (eq_attr "cpu" "znver1,znver2")
(and (eq_attr "type" "idiv")
(and (eq_attr "mode" "QI")
(eq_attr "memory" "none"))))
- "znver1-direct,znver1-ieu2*12")
+ "znver1-direct,znver1-idiv*13")
;; Mem operands
(define_insn_reservation "znver1_idiv_mem_DI" 45
@@ -205,84 +208,84 @@ (define_insn_reservation "znver1_idiv_mem_DI" 45
(and (eq_attr "type" "idiv")
(and (eq_attr "mode" "DI")
(eq_attr "memory" "none"))))
- "znver1-double,znver1-load,znver1-ieu2*41")
+ "znver1-double,znver1-load,znver1-idiv*14")
(define_insn_reservation "znver1_idiv_mem_SI" 29
(and (eq_attr "cpu" "znver1,znver2")
(and (eq_attr "type" "idiv")
(and (eq_attr "mode" "SI")
(eq_attr "memory" "none"))))
- "znver1-double,znver1-load,znver1-ieu2*25")
+ "znver1-double,znver1-load,znver1-idiv*14")
(define_insn_reservation "znver1_idiv_mem_HI" 21
(and (eq_attr "cpu" "znver1,znver2")
(and (eq_attr "type" "idiv")
(and (eq_attr "mode" "HI")
(eq_attr "memory" "none"))))
- "znver1-double,znver1-load,znver1-ieu2*17")
+ "znver1-double,znver1-load,znver1-idiv*14")
(define_insn_reservation "znver1_idiv_mem_QI" 16
(and (eq_attr "cpu" "znver1,znver2")
(and (eq_attr "type" "idiv")
(and (eq_attr "mode" "QI")
(eq_attr "memory" "none"))))
- "znver1-direct,znver1-load,znver1-ieu2*12")
+ "znver1-direct,znver1-load,znver1-idiv*13")
(define_insn_reservation "znver3_idiv_DI" 18
(and (eq_attr "cpu" "znver3")
(and (eq_attr "type" "idiv")
(and (eq_attr "mode" "DI")
(eq_attr "memory" "none"))))
- "znver1-double,znver1-ieu2*18")
+ "znver1-double,znver1-idiv*7")
(define_insn_reservation "znver3_idiv_SI" 12
(and (eq_attr "cpu" "znver3")
(and (eq_attr "type" "idiv")
(and (eq_attr "mode" "SI")
(eq_attr "memory" "none"))))
- "znver1-double,znver1-ieu2*12")
+ "znver1-double,znver1-idiv*6")
(define_insn_reservation "znver3_idiv_HI" 10
(and (eq_attr "cpu" "znver3")
(and (eq_attr "type" "idiv")
(and (eq_attr "mode" "HI")
(eq_attr "memory" "none"))))
- "znver1-double,znver1-ieu2*10")
+ "znver1-double,znver1-idiv*4")
(define_insn_reservation "znver3_idiv_QI" 9
(and (eq_attr "cpu" "znver3")
(and (eq_attr "type" "idiv")
(and (eq_attr "mode" "QI")
(eq_attr "memory" "none"))))
- "znver1-direct,znver1-ieu2*9")
+ "znver1-direct,znver1-idiv*4")
(define_insn_reservation "znver3_idiv_mem_DI" 22
(and (eq_attr "cpu" "znver3")
(and (eq_attr "type" "idiv")
(and (eq_attr "mode" "DI")
(eq_attr "memory" "load"))))
- "znver1-double,znver1-load,znver1-ieu2*22")
+ "znver1-double,znver1-load,znver1-idiv*7")
(define_insn_reservation "znver3_idiv_mem_SI" 16
(and (eq_attr "cpu" "znver3")
(and (eq_attr "type" "idiv")
(and (eq_attr "mode" "SI")
(eq_attr "memory" "load"))))
- "znver1-double,znver1-load,znver1-ieu2*16")
+ "znver1-double,znver1-load,znver1-idiv*6")
(define_insn_reservation "znver3_idiv_mem_HI" 14
(and (eq_attr "cpu" "znver3")
(and (eq_attr "type" "idiv")
(and (eq_attr "mode" "HI")
(eq_attr "memory" "load"))))
- "znver1-double,znver1-load,znver1-ieu2*10")
+ "znver1-double,znver1-load,znver1-idiv*4")
(define_insn_reservation "znver3_idiv_mem_QI" 13
(and (eq_attr "cpu" "znver3")
(and (eq_attr "type" "idiv")
(and (eq_attr "mode" "QI")
(eq_attr "memory" "load"))))
- "znver1-direct,znver1-load,znver1-ieu2*9")
+ "znver1-direct,znver1-load,znver1-idiv*4")
;; STR ISHIFT which are micro coded.
;; Fix me: Latency need to be rechecked.
--
2.37.2
