q3aql/Serious-Engine
q3aql/Serious-Engine
1
0
Fork 0
mirror of https://github.com/ptitSeb/Serious-Engine synced 2024-11-25 03:40:26 +01:00
Code Issues Packages Projects Releases Wiki Activity

fix SoundMixer asm

Browse Source 
SoundMixer386.asm forgot to save some callee-save registers too
This commit is contained in:
notaz 2016-04-22 02:25:24 +03:00
parent b1837c2e58
commit 52c178718c
2 changed files with 24 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

24
Sources/Engine/Sound/SoundMixer.cpp
View File

@ -96,11 +96,12 @@ void ResetMixer( const SLONG *pslBuffer, const SLONG slBufferSize)
} }
#elif (defined __GNU_INLINE__) #elif (defined __GNU_INLINE__)
// !!! FIXME : rcg12172001 Is this REALLY any faster than memset()? // !!! FIXME : rcg12172001 Is this REALLY any faster than memset()?
ULONG clob1, clob2;
__asm__ __volatile__ ( __asm__ __volatile__ (
"cld \n\t" "cld \n\t"
"rep \n\t" "rep \n\t"
"stosl \n\t" "stosl \n\t"
: // no outputs. : "=D" (clob1), "=c" (clob2)
: "a" (0), "D" (pvMixerBuffer), "c" (slMixerBufferSize*2) : "a" (0), "D" (pvMixerBuffer), "c" (slMixerBufferSize*2)
: "cc", "memory" : "cc", "memory"
); );
@ -132,11 +133,12 @@ void CopyMixerBuffer_stereo( const SLONG slSrcOffset, void *pDstBuffer, const SL
} }
#elif (defined __GNU_INLINE__) #elif (defined __GNU_INLINE__)
// !!! FIXME : rcg12172001 Is this REALLY any faster than memcpy()? // !!! FIXME : rcg12172001 Is this REALLY any faster than memcpy()?
ULONG clob1, clob2, clob3;
__asm__ __volatile__ ( __asm__ __volatile__ (
"cld \n\t" "cld \n\t"
"rep \n\t" "rep \n\t"
"movsl \n\t" "movsl \n\t"
: // no outputs. : "=S" (clob1), "=D" (clob2), "=c" (clob3)
: "S" (((char *)pvMixerBuffer) + slSrcOffset), : "S" (((char *)pvMixerBuffer) + slSrcOffset),
"D" (pDstBuffer), "D" (pDstBuffer),
"c" (slBytes >> 2) "c" (slBytes >> 2)
@ -184,6 +186,9 @@ copyLoop:
#elif (defined __GNU_INLINE__) #elif (defined __GNU_INLINE__)
__asm__ __volatile__ ( __asm__ __volatile__ (
"movl %[pvMixerBuffer], %%esi \n\t"
"movl %[pDstBuffer], %%edi \n\t"
"movl %[slDW], %%ecx \n\t"
"0: \n\t" // copyLoop "0: \n\t" // copyLoop
"movzwl (%%esi), %%eax \n\t" "movzwl (%%esi), %%eax \n\t"
"movw %%ax, (%%edi) \n\t" "movw %%ax, (%%edi) \n\t"
@ -192,10 +197,10 @@ copyLoop:
"decl %%ecx \n\t" "decl %%ecx \n\t"
"jnz 0b \n\t" // copyLoop "jnz 0b \n\t" // copyLoop
: // no outputs. : // no outputs.
: "S" (((char *)pvMixerBuffer) + slSrcOffset), : [pvMixerBuffer] "g" (((char *)pvMixerBuffer) + slSrcOffset),
"D" (pDstBuffer), [pDstBuffer] "g" (pDstBuffer),
"c" (slBytes >> 2) [slDW] "g" (slBytes >> 2)
: "cc", "memory", "eax" : "eax", "ecx", "esi", "edi", "cc", "memory"
); );
#else #else
@ -247,6 +252,9 @@ copyLoop:
#elif (defined __GNU_INLINE__) #elif (defined __GNU_INLINE__)
__asm__ __volatile__ ( __asm__ __volatile__ (
"movl %[pvMixerBuffer], %%esi \n\t"
"movl %[pvMixerBuffer], %%edi \n\t"
"movl %[slDW], %%ecx \n\t"
"cld \n\t" "cld \n\t"
"0: \n\t" // copyLoop "0: \n\t" // copyLoop
"movq (%%esi), %%mm0 \n\t" "movq (%%esi), %%mm0 \n\t"
@ -258,8 +266,8 @@ copyLoop:
"jnz 0b \n\t" // copyLoop "jnz 0b \n\t" // copyLoop
"emms \n\t" "emms \n\t"
: // no outputs. : // no outputs.
: "S" (pvMixerBuffer), "D" (pvMixerBuffer), "c" (slBytes >> 2) : [pvMixerBuffer] "g" (pvMixerBuffer), [slDW] "g" (slBytes >> 2)
: "cc", "memory" : FPU_REGS, "mm0", "ecx", "esi", "edi", "cc", "memory"
); );
#else #else

8
Sources/Engine/Sound/SoundMixer386.asm
View File

@ -96,6 +96,8 @@ SEGMENT .text
global MixMono_asm global MixMono_asm
MixMono_asm: MixMono_asm:
push ebx ; Save GCC register. push ebx ; Save GCC register.
push esi
push edi
; convert from floats to fixints 32:16 ; convert from floats to fixints 32:16
fld D [fLeftOfs] fld D [fLeftOfs]
fmul D [f65536] fmul D [f65536]
@ -224,6 +226,8 @@ loopEnd_MixMono:
shr edx,16 shr edx,16
mov D [slLastLeftSample],eax mov D [slLastLeftSample],eax
mov D [slLastRightSample],edx mov D [slLastRightSample],edx
pop edi
pop esi
pop ebx ; Restore GCC register. pop ebx ; Restore GCC register.
emms emms
ret ret
@ -232,6 +236,8 @@ loopEnd_MixMono:
global MixStereo_asm global MixStereo_asm
MixStereo_asm: MixStereo_asm:
push ebx ; Save GCC register. push ebx ; Save GCC register.
push esi
push edi
; convert from floats to fixints 32:16 ; convert from floats to fixints 32:16
fld D [fLeftOfs] fld D [fLeftOfs]
fmul D [f65536] fmul D [f65536]
@ -363,6 +369,8 @@ loopEnd_MixStereo:
mov D [slLastLeftSample],eax mov D [slLastLeftSample],eax
mov D [slLastRightSample],edx mov D [slLastRightSample],edx
emms emms
pop edi
pop esi
pop ebx ; Restore GCC register. pop ebx ; Restore GCC register.
ret ret