mirror/sointu
1
0
Fork 0
mirror of https://github.com/vsariola/sointu.git synced 2025-06-04 01:28:45 -04:00
Code Issues Packages Projects Releases Wiki Activity

Optimize delay for better register use.

Browse Source
This commit is contained in:
Veikko Sariola 2020-05-23 17:40:22 +03:00
parent da52c10f7f
commit 5c25eacd08
1 changed files with 15 additions and 19 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

34
src/opcodes/effects.asm
View File

@ -315,15 +315,11 @@ SECT_TEXT(sudelay)
EXPORT MANGLE_FUNC(su_op_delay,0) EXPORT MANGLE_FUNC(su_op_delay,0)
lodsw ; al = delay index, ah = delay count lodsw ; al = delay index, ah = delay count
push_registers VAL, COM, WRK ; these are non-volatile according to our convention push_registers VAL, COM ; these are non-volatile according to our convention
%ifdef INCLUDE_DELAY_MODULATION ; TODO: at the moment, this must be manually enabled, as the PORT macro is a singleline macro and cannot toggle this. Will have to think of a good solution.
mov esi, dword [WRK+su_unit.ports+su_delay_ports.delaymod]
mov [INP+su_delay_ports.delaymod],esi ; move the modulation to the transformed values, as we will not have access to WRK soon
%endif
movzx ebx, al movzx ebx, al
apply {lea _BX,},MANGLE_DATA(su_delay_times),_BX*2,{} ; _BP now points to the right position within delay time table apply {lea _BX,},MANGLE_DATA(su_delay_times),_BX*2,{} ; _BP now points to the right position within delay time table
movzx esi, word [_SP + su_stack.tick + PUSH_REG_SIZE(3)] ; notice that we load word, so we wrap at 65536 movzx esi, word [_SP + su_stack.tick + PUSH_REG_SIZE(2)] ; notice that we load word, so we wrap at 65536
mov WRK, PTRWORD [_SP + su_stack.delaywrk + PUSH_REG_SIZE(3)] ; WRK is now the separate delay workspace, as they require a lot more space mov _CX, PTRWORD [_SP + su_stack.delaywrk + PUSH_REG_SIZE(2)] ; WRK is now the separate delay workspace, as they require a lot more space
%ifdef INCLUDE_STEREO_DELAY %ifdef INCLUDE_STEREO_DELAY
jnc su_op_delay_mono jnc su_op_delay_mono
push _AX ; save _ah (delay count) push _AX ; save _ah (delay count)
@ -334,8 +330,8 @@ EXPORT MANGLE_FUNC(su_op_delay,0)
su_op_delay_mono: ; flow into mono delay su_op_delay_mono: ; flow into mono delay
%endif %endif
call su_op_delay_do ; when stereo delay is not enabled, we could inline this to save 5 bytes, but I expect stereo delay to be farely popular so maybe not worth the hassle call su_op_delay_do ; when stereo delay is not enabled, we could inline this to save 5 bytes, but I expect stereo delay to be farely popular so maybe not worth the hassle
mov PTRWORD [_SP + su_stack.delaywrk + PUSH_REG_SIZE(3)],WRK ; move delay workspace pointer back to stack. mov PTRWORD [_SP + su_stack.delaywrk + PUSH_REG_SIZE(2)],_CX ; move delay workspace pointer back to stack.
pop_registers VAL, COM, WRK pop_registers VAL, COM
%ifdef INCLUDE_DELAY_MODULATION %ifdef INCLUDE_DELAY_MODULATION
xor eax, eax xor eax, eax
mov dword [WRK+su_unit.ports+su_delay_ports.delaymod], eax ; zero it mov dword [WRK+su_unit.ports+su_delay_ports.delaymod], eax ; zero it
@ -377,7 +373,7 @@ su_op_delay_loop:
su_op_delay_skipnotesync: su_op_delay_skipnotesync:
%endif %endif
%ifdef INCLUDE_DELAY_MODULATION %ifdef INCLUDE_DELAY_MODULATION
fld dword [INP+su_delay_ports.delaymod] fld dword [WRK+su_unit.ports+su_delay_ports.delaymod]
apply fmul dword, c_32767 ; scale it up, as the modulations would be too small otherwise apply fmul dword, c_32767 ; scale it up, as the modulations would be too small otherwise
faddp st1, st0 faddp st1, st0
%endif %endif
@ -388,33 +384,33 @@ su_op_delay_loop:
mov edi, esi mov edi, esi
sub di, word [_BX] ; we perform the math in 16-bit to wrap around sub di, word [_BX] ; we perform the math in 16-bit to wrap around
%endif %endif
fld dword [WRK+su_delayline_wrk.buffer+_DI*4]; s dr*y p*p*x, where s is the sample from delay buffer fld dword [_CX+su_delayline_wrk.buffer+_DI*4]; s dr*y p*p*x, where s is the sample from delay buffer
fadd st1, st0 ; s dr*y+s p*p*x (add comb output to current output) fadd st1, st0 ; s dr*y+s p*p*x (add comb output to current output)
fld1 ; 1 s dr*y+s p*p*x fld1 ; 1 s dr*y+s p*p*x
fsub dword [INP+su_delay_ports.damp] ; 1-da s dr*y+s p*p*x fsub dword [INP+su_delay_ports.damp] ; 1-da s dr*y+s p*p*x
fmulp st1, st0 ; s*(1-da) dr*y+s p*p*x fmulp st1, st0 ; s*(1-da) dr*y+s p*p*x
fld dword [INP+su_delay_ports.damp] ; da s*(1-da) dr*y+s p*p*x fld dword [INP+su_delay_ports.damp] ; da s*(1-da) dr*y+s p*p*x
fmul dword [WRK+su_delayline_wrk.filtstate] ; o*da s*(1-da) dr*y+s p*p*x, where o is stored fmul dword [_CX+su_delayline_wrk.filtstate] ; o*da s*(1-da) dr*y+s p*p*x, where o is stored
faddp st1, st0 ; o*da+s*(1-da) dr*y+s p*p*x faddp st1, st0 ; o*da+s*(1-da) dr*y+s p*p*x
fst dword [WRK+su_delayline_wrk.filtstate] ; o'=o*da+s*(1-da), o' dr*y+s p*p*x fst dword [_CX+su_delayline_wrk.filtstate] ; o'=o*da+s*(1-da), o' dr*y+s p*p*x
fmul dword [INP+su_delay_ports.feedback] ; f*o' dr*y+s p*p*x fmul dword [INP+su_delay_ports.feedback] ; f*o' dr*y+s p*p*x
fadd st0, st2 ; f*o'+p*p*x dr*y+s p*p*x fadd st0, st2 ; f*o'+p*p*x dr*y+s p*p*x
fstp dword [WRK+su_delayline_wrk.buffer+_SI*4]; save f*o'+p*p*x to delay buffer fstp dword [_CX+su_delayline_wrk.buffer+_SI*4]; save f*o'+p*p*x to delay buffer
add _BX,2 ; move to next index add _BX,2 ; move to next index
add WRK, su_delayline_wrk.size ; go to next delay delay workspace add _CX, su_delayline_wrk.size ; go to next delay delay workspace
sub ah, 2 sub ah, 2
jg su_op_delay_loop ; if ah > 0, goto loop jg su_op_delay_loop ; if ah > 0, goto loop
fstp st1 ; dr*y+s1+s2+s3+... fstp st1 ; dr*y+s1+s2+s3+...
; DC-filtering ; DC-filtering
fld dword [WRK+su_delayline_wrk.dcout] ; o s fld dword [_CX+su_delayline_wrk.dcout] ; o s
apply fmul dword, c_dc_const ; c*o s apply fmul dword, c_dc_const ; c*o s
fsub dword [WRK+su_delayline_wrk.dcin] ; c*o-i s fsub dword [_CX+su_delayline_wrk.dcin] ; c*o-i s
fxch ; s c*o-i fxch ; s c*o-i
fst dword [WRK+su_delayline_wrk.dcin] ; i'=s, s c*o-i fst dword [_CX+su_delayline_wrk.dcin] ; i'=s, s c*o-i
faddp st1 ; s+c*o-i faddp st1 ; s+c*o-i
apply fadd dword, c_0_5 ; add and sub small offset to prevent denormalization apply fadd dword, c_0_5 ; add and sub small offset to prevent denormalization
apply fsub dword, c_0_5 apply fsub dword, c_0_5
fst dword [WRK+su_delayline_wrk.dcout] ; o'=s+c*o-i fst dword [_CX+su_delayline_wrk.dcout] ; o'=s+c*o-i
ret ret
;------------------------------------------------------------------------------- ;-------------------------------------------------------------------------------