📄 dsk_app.asm
字号:
;** --------------------------------------------------------------------------*
NOP 2
MV .S1X B7,A6
MVC .S2 B3,CSR ; interrupts on
MV .S1X B5,A5
.line 16
LDW .D1T1 *A10++,A3 ; |893|
|| LDW .D2T2 *B10++,B4 ; |893|
|| ADDSP .L1 A6,A4,A4 ; |893|
NOP 3
ADDSP .L1 A5,A4,A4 ; |893|
NOP 2
MPYSP .M1 A12,A3,A3 ; |893|
|| MPYSP .M2 B11,B4,B4 ; |893|
ADDSP .L1 A0,A4,A0 ; |893|
NOP 3
ADDSP .L1X B4,A0,A0 ; |893|
NOP 3
ADDSP .L1 A3,A0,A0 ; |893|
NOP 3
CMPGTSP .S1 A0,A9,A1 ; |893|
.line 17
[ A1] MV .D1 A0,A9 ; |894|
.line 18
[ A1] MV .D1 A8,A11 ; |895|
.line 20
ADD .D1 1,A8,A8 ; |897|
|| ADD .D2 4,B9,B9 ; |897|
|| SUB .S2 B2,1,B2 ; |897|
[ B2] B .S2 L1 ; |897|
NOP 5
; BRANCH OCCURS ; |897|
;** --------------------------------------------------------------------------*
.line 21
MV .D1 A11,A4 ; |898|
.line 22
LDW .D2T2 *+SP(20),B3 ; |899|
MV .S1X SP,A9 ; |899|
LDW .D1T1 *+A9(4),A12 ; |899|
LDW .D1T1 *+A9(16),A13 ; |899|
NOP 1
RET .S2 B3 ; |899|
|| LDDW .D2T2 *+SP(24),B11:B10 ; |899|
LDW .D2T2 *++SP(32),B12 ; |899|
|| LDDW .D1T1 *+A9(8),A11:A10 ; |899|
NOP 4
; BRANCH OCCURS ; |899|
.endfunc 899,01c083c00h,32
.sect ".text"
.global _trainingRun
.sym _trainingRun,_trainingRun, 32, 2, 0
.func 803
;******************************************************************************
;* FUNCTION NAME: _trainingRun *
;* *
;* Regs Modified : A0,A1,A2,A3,A4,A5,A6,A7,A8,A9,A10,B0,B1,B2,B3,B4,B5, *
;* B6,B7,B8,B9,B10,B11,SP *
;* Regs Used : A0,A1,A2,A3,A4,A5,A6,A7,A8,A9,A10,B0,B1,B2,B3,B4,B5, *
;* B6,B7,B8,B9,B10,B11,DP,SP *
;* Local Frame Size : 0 Args + 0 Auto + 16 Save = 16 byte *
;******************************************************************************
;******************************************************************************
;* *
;* Using -g (debug) with optimization (-o3) may disable key optimizations! *
;* *
;******************************************************************************
_trainingRun:
;** --------------------------------------------------------------------------*
.line 2
.sym _framebuffer,4, 19, 17, 32
.sym _framebuffer,10, 19, 4, 32
STW .D2T2 B11,*SP--(16) ; |804|
STW .D2T2 B10,*+SP(12) ; |804|
STW .D2T2 B3,*+SP(8) ; |804|
STW .D2T1 A10,*+SP(4) ; |804|
MV .D1 A4,A10 ; |804|
.line 7
MVKL .S1 _memset,A0 ; |809|
MVKH .S1 _memset,A0 ; |809|
;*----------------------------------------------------------------------------*
;* SOFTWARE PIPELINE INFORMATION
;*
;* Loop source line : 811
;* Loop opening brace source line : 812
;* Loop closing brace source line : 812
;* Loop Unroll Multiple : 2x
;* Known Minimum Trip Count : 1192
;* Known Maximum Trip Count : 1192
;* Known Max Trip Count Factor : 1192
;* Loop Carried Dependency Bound(^) : 0
;* Unpartitioned Resource Bound : 2
;* Partitioned Resource Bound(*) : 2
;* Resource Partition:
;* A-side B-side
;* .L units 0 2*
;* .S units 1 0
;* .D units 2* 2*
;* .M units 0 0
;* .X cross paths 0 2*
;* .T address paths 2* 2*
;* Long read paths 0 2*
;* Long write paths 0 0
;* Logical ops (.LS) 0 0 (.L or .S unit)
;* Addition ops (.LSD) 1 0 (.L or .S or .D unit)
;* Bound(.L .S .LS) 1 1
;* Bound(.L .S .D .LS .LSD) 2* 2*
;*
;* Searching for software pipeline schedule at ...
;* ii = 2 Schedule found with 6 iterations in parallel
;* Done
;*
;* Epilog not removed
;* Collapsed epilog stages : 0
;* Collapsed prolog stages : 5
;* Minimum required memory pad : 0 bytes
;*
;* For further improvement on this loop, try option -mh20
;*
;* Minimum safe trip count : 5 (after unrolling)
;*----------------------------------------------------------------------------*
L5: ; PIPED LOOP PROLOG
CALL .S2X A0 ; |809|
NOP 2
MVK .S1 0x27c0,A6 ; |809|
MVKL .S1 _temp,A4 ; |809|
|| MVKL .S2 RL0,B3 ; |809|
ZERO .D2 B4 ; |809|
|| MVKH .S2 RL0,B3 ; |809|
|| MVKH .S1 _temp,A4 ; |809|
RL0: ; CALL OCCURS ; |809|
NOP 1
MVKL .S2 _temp+320,B6
MVKL .S2 _temp+324,B4
MVKH .S2 _temp+324,B4
SUB .D1 A10,4,A0
|| MVKH .S2 _temp+320,B6
.line 9
MVC .S2 CSR,B7
AND .S2 -2,B7,B5
MVC .S2 B5,CSR ; interrupts off
B .S1 L6 ; |812| (P) <0,5>
MVK .S1 0x4a8,A3 ; |811|
B .S1 L6 ; |812| (P) <1,5>
MVK .S1 0x5,A2 ; init prolog collapse predicate
|| MVK .S2 0x1,B0 ; init prolog collapse predicate
|| SUB .D1 A3,3,A1
;** --------------------------------------------------------------------------*
L6: ; PIPED LOOP KERNEL
[!A2] STW .D2T2 B5,*B6++(8) ; |812| <0,9>
|| INTSP .L2X A3,B5 ; |812| <2,5>
|| [ A1] B .S1 L6 ; |812| <2,5>
|| [!B0] LDH .D1T1 *+A0(2),A3 ; |812| <4,1>
[ B0] SUB .S2 B0,1,B0 ; <0,10>
|| [ A2] SUB .S1 A2,1,A2 ; <0,10>
|| [!A2] STW .D2T2 B5,*B4++(8) ; |812| <0,10>
|| INTSP .L2X A3,B5 ; |812| <2,6>
|| [ A1] SUB .L1 A1,1,A1 ; |812| <3,4>
|| LDH .D1T1 *++A0(4),A3 ; |812| <5,0>
;*----------------------------------------------------------------------------*
;* SOFTWARE PIPELINE INFORMATION
;*
;* Loop source line : 818
;* Loop opening brace source line : 819
;* Loop closing brace source line : 822
;* Loop Unroll Multiple : 2x
;* Known Minimum Trip Count : 128
;* Known Maximum Trip Count : 128
;* Known Max Trip Count Factor : 128
;* Loop Carried Dependency Bound(^) : 0
;* Unpartitioned Resource Bound : 3
;* Partitioned Resource Bound(*) : 3
;* Resource Partition:
;* A-side B-side
;* .L units 0 0
;* .S units 1 0
;* .D units 3* 3*
;* .M units 0 0
;* .X cross paths 0 0
;* .T address paths 3* 3*
;* Long read paths 2 2
;* Long write paths 0 0
;* Logical ops (.LS) 0 0 (.L or .S unit)
;* Addition ops (.LSD) 0 1 (.L or .S or .D unit)
;* Bound(.L .S .LS) 1 0
;* Bound(.L .S .D .LS .LSD) 2 2
;*
;* Searching for software pipeline schedule at ...
;* ii = 3 Schedule found with 3 iterations in parallel
;* Done
;*
;* Collapsed epilog stages : 2
;* Prolog not entirely removed
;* Collapsed prolog stages : 1
;*
;* Minimum required memory pad : 0 bytes
;*
;* For further improvement on this loop, try option -mh128
;*
;* Minimum safe trip count : 1 (after unrolling)
;*----------------------------------------------------------------------------*
L7: ; PIPED LOOP EPILOG AND PROLOG
INTSP .L2X A3,B5 ; |812| (E) <3,5>
|| LDH .D1T1 *+A0(2),A3 ; |812| (E) <5,1>
|| STW .D2T2 B5,*B6++(8) ; |812| (E) <1,9>
STW .D2T2 B5,*B4++(8) ; |812| (E) <1,10>
|| INTSP .L2X A3,B5 ; |812| (E) <3,6>
INTSP .L2X A3,B5 ; |812| (E) <4,5>
|| STW .D2T2 B5,*B6++(8) ; |812| (E) <2,9>
STW .D2T2 B5,*B4++(8) ; |812| (E) <2,10>
|| INTSP .L2X A3,B5 ; |812| (E) <4,6>
INTSP .L2X A3,B5 ; |812| (E) <5,5>
|| STW .D2T2 B5,*B6++(8) ; |812| (E) <3,9>
STW .D2T2 B5,*B4++(8) ; |812| (E) <3,10>
|| INTSP .L2X A3,B5 ; |812| (E) <5,6>
STW .D2T2 B5,*B6++(8) ; |812| (E) <4,9>
STW .D2T2 B5,*B4++(8) ; |812| (E) <4,10>
MVC .S2 B7,CSR ; interrupts on
|| STW .D2T2 B5,*B6++(8) ; |812| (E) <5,9>
STW .D2T2 B5,*B4++(8) ; |812| (E) <5,10>
.line 7
MVKL .S1 _temp,A0 ; |809|
MVKH .S1 _temp,A0 ; |809|
MV .S2X A0,B10 ; |809|
.line 12
MV .L1X B10,A4 ; |814|
|| MVKL .S1 _rx,A10 ; |814|
|| MVKL .S2 _DSPF_sp_convol,B5 ; |814|
MVKH .S1 _rx,A10 ; |814|
|| MVKH .S2 _DSPF_sp_convol,B5 ; |814|
MVK .S1 0x9a0,A8 ; |814|
|| MV .D1 A10,A6 ; |814|
|| CALL .S2 B5 ; |814|
MVKL .S2 _cos_filter,B4 ; |814|
MVKL .S2 RL1,B3 ; |814|
MVKH .S2 RL1,B3 ; |814|
MVKH .S2 _cos_filter,B4 ; |814|
MVK .S2 0x50,B6 ; |814|
RL1: ; CALL OCCURS ; |814|
.line 14
NOP 1
MV .D1 A10,A4 ; |816|
|| MVKL .S2 RL2,B3 ; |816|
|| MVKL .S1 _sync,A0 ; |816|
MVKL .S2 _training,B4 ; |816|
|| MVKH .S1 _sync,A0 ; |816|
CALL .S2X A0 ; |816|
MVKH .S2 _training,B4 ; |816|
MVKH .S2 RL2,B3 ; |816|
NOP 3
RL2: ; CALL OCCURS ; |816|
NOP 1
SUB .D2 B10,16,B7
|| EXT .S1 A4,16,16,A0 ; |816|
|| MVKL .S2 _RefIndex,B4 ; |816|
MVK .S1 1024,A3
|| ADDAW .D1 A10,A0,A0
|| MVKH .S2 _RefIndex,B4 ; |816|
ADD .D1 A3,A0,A4
|| MVK .S2 1056,B4
|| STH .D2T1 A4,*B4 ; |816|
ZERO .D2 B4
|| ADD .S2X B4,A0,B6
.line 16
MV .L1X B4,A3
|| MVK .S2 0x80,B5 ; |818|
ADD .D2 4,B7,B7
|| LDW .D1T1 *A4++(64),A5 ; |820| (P) <0,0>
|| B .S1 L8 ; |822| (P) <0,3>
|| MV .L1X B7,A0
|| MVK .S2 0x1,B1 ; init prolog collapse predicate
|| SUB .L2 B5,1,B0
SUB .S1X B5,1,A1
ADD .S1X 1,B0,A2
;** --------------------------------------------------------------------------*
L8: ; PIPED LOOP KERNEL
[ A2] LDW .D2T2 *B6++(64),B5 ; |820| <1,3>
|| [ B0] B .S1 L8 ; |822| <1,3>
|| [ A1] LDW .D1T1 *A4++(64),A5 ; |820| <2,0>
[!B1] STW .D2T2 B4,*++B7(16) ; |821| <0,7>
|| [!B1] ST⌨️ 快捷键说明
复制代码
Ctrl + C
搜索代码
Ctrl + F
全屏模式
F11
切换主题
Ctrl + Shift + D
显示快捷键
?
增大字号
Ctrl + =
减小字号
Ctrl + -