ysimard/be_chti
1
1
Fork 0
Code Issues Pull requests Releases Wiki Activity

Amélioration de la vitesse

Browse source
This commit is contained in:
Yohan Simard 2020-04-28 15:59:54 +02:00
parent a047b8cb7d
commit 5868ca0b18
2 changed files with 54 additions and 40 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

64
challenge/Project.uvoptx
View file

@ -157,38 +157,6 @@
<Bp>
<Number>0</Number>
<Type>0</Type>
<LineNumber>38</LineNumber>
<EnabledFlag>1</EnabledFlag>
<Address>134218234</Address>
<ByteObject>0</ByteObject>
<HtxType>0</HtxType>
<ManyObjects>0</ManyObjects>
<SizeOfObject>0</SizeOfObject>
<BreakByAccess>0</BreakByAccess>
<BreakIfRCount>1</BreakIfRCount>
<Filename>.\Src\calcul_dft.s</Filename>
<ExecCommand></ExecCommand>
<Expression>\\CHTI\Src/calcul_dft.s\38</Expression>
</Bp>
<Bp>
<Number>1</Number>
<Type>0</Type>
<LineNumber>0</LineNumber>
<EnabledFlag>0</EnabledFlag>
<Address>134218136</Address>
<ByteObject>0</ByteObject>
<HtxType>0</HtxType>
<ManyObjects>0</ManyObjects>
<SizeOfObject>0</SizeOfObject>
<BreakByAccess>0</BreakByAccess>
<BreakIfRCount>1</BreakIfRCount>
<Filename></Filename>
<ExecCommand></ExecCommand>
<Expression>0x08000198</Expression>
</Bp>
<Bp>
<Number>2</Number>
<Type>0</Type>
<LineNumber>10</LineNumber>
<EnabledFlag>1</EnabledFlag>
<Address>134218152</Address>
@ -202,6 +170,38 @@
<ExecCommand></ExecCommand>
<Expression>\\CHTI\Src/principal.c\10</Expression>
</Bp>
<Bp>
<Number>1</Number>
<Type>0</Type>
<LineNumber>7</LineNumber>
<EnabledFlag>1</EnabledFlag>
<Address>134218128</Address>
<ByteObject>0</ByteObject>
<HtxType>0</HtxType>
<ManyObjects>0</ManyObjects>
<SizeOfObject>0</SizeOfObject>
<BreakByAccess>0</BreakByAccess>
<BreakIfRCount>1</BreakIfRCount>
<Filename>.\Src\principal.c</Filename>
<ExecCommand></ExecCommand>
<Expression>\\CHTI\Src/principal.c\7</Expression>
</Bp>
<Bp>
<Number>2</Number>
<Type>0</Type>
<LineNumber>52</LineNumber>
<EnabledFlag>1</EnabledFlag>
<Address>134218270</Address>
<ByteObject>0</ByteObject>
<HtxType>0</HtxType>
<ManyObjects>0</ManyObjects>
<SizeOfObject>0</SizeOfObject>
<BreakByAccess>0</BreakByAccess>
<BreakIfRCount>1</BreakIfRCount>
<Filename>.\Src\calcul_dft.s</Filename>
<ExecCommand></ExecCommand>
<Expression>\\CHTI\Src/calcul_dft.s\52</Expression>
</Bp>
<Bp>
<Number>3</Number>
<Type>0</Type>

30
challenge/Src/calcul_dft.s
View file

@ -4,34 +4,48 @@
N equ 64
; vitesse initiale : 1726
; vitesse actuelle : 1119
; vitesse actuelle : 1101
dft proc
; r0 = x (signal), r1 = k
push {r4, r5, r6, r7, r8, r9}
ldr r2, =TabCos
ldr r3, =TabSin
mov r4, #0 ; r4 = i
mov r5, #0 ; r5 = ik
mov r9, #0 ; r9 = résultat réel
mov r12, #0 ; r10 = résultat imaginaire
; première itération de la boucle
ldrh r6, [r0], #2 ; r6 = x[i] (puis mise à jour de r0) | format 4.12
mov r7, #32767 ; r7 = cos [ik%N] | format 1.15
mov r8, #0 ; r8 = sin [ik%N] | format 1.15
mul r9, r6, r7 ; r9 = x[i] * cos[ik%n] | format 5.27
mul r12, r6, r8 ; r12 = x[i] * sin[ik%n] | format 5.27
add r5, r1 ; r5 = (i-1)k + k
and r5, #N-1 ; r5 = r5 % 64
mov r4, #1 ; r4 = i = 1
deb_somme ldrh r6, [r0, r4, LSL #1] ; r6 = x[i], LSL #1 pour faire ×2 car chaque valeur est sur 2 octets | format 4.12
deb_somme ldrh r6, [r0], #2 ; r6 = x[i] (puis mise à jour de r0) | format 4.12
ldrsh r7, [r2, r5, LSL #1] ; r7 = cos [ik%N] | format 1.15
ldrsh r8, [r3, r5, LSL #1] ; r8 = sin [ik%N] | format 1.15
mla r9, r6, r7, r9 ; r9 += x[i] * cos[ik%n] | format 5.27
mla r12, r6, r8, r12 ; r10 += x[i] * sin[ik%n] | format 5.27
mla r12, r6, r8, r12 ; r12 += x[i] * sin[ik%n] | format 5.27
add r5, r1 ; r5 = (i-1)k + k
and r5, #N-1 ; r5 = r5 % 64
add r4, #1 ; on incrémente i
cmp r4, #N ;
cmp r4, #N-1
bne deb_somme
; dernière itération de la boucle (pour éviter les 4 instructions inutiles à la fin)
ldrh r6, [r0] ; r6 = x[i] | format 4.12
ldrsh r7, [r2, r5, LSL #1] ; r7 = cos [ik%N] | format 1.15
ldrsh r8, [r3, r5, LSL #1] ; r8 = sin [ik%N] | format 1.15
mla r9, r6, r7, r9 ; r9 += x[i] * cos[ik%n] | format 5.27
mla r12, r6, r8, r12 ; r10 += x[i] * sin[ik%n] | format 5.27
; etat des registres : r0 = x, r1 = k, r9 = Re, r10 = Im, autres libres
smull r2, r0, r9, r9 ; (r2,r0) = Re(k)² | format : 10.54
smlal r2, r0, r12, r12 ; (r2,r0) = Re(k)² + Im(k)² | format : 10.54
pop {r4, r5, r6, r7, r8, r9}