mirrors/pulseaudio
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/pulseaudio/pulseaudio.git synced 2025-11-06 13:29:56 -05:00
Code Issues Projects Releases Packages Wiki Activity Actions

remap: fix counters for mmx and sse remap

Browse source 
Take the size of the sample into account when calculating the amount of samples
we process in parallel.
This commit is contained in:
Wim Taymans 2009-08-29 12:22:42 +02:00 • committed by Lennart Poettering
parent dee2aa3f05
commit dc221f204b
2 changed files with 42 additions and 26 deletions
Download patch file Download diff file Expand all files Collapse all files

34
src/pulsecore/remap_mmx.c
View file

@ -65,16 +65,24 @@
" add $32, %1 \n\t" \ " add $32, %1 \n\t" \
" add $64, %0 \n\t" " add $64, %0 \n\t"
#define HANDLE_SINGLE(s) \ #define HANDLE_SINGLE_dq() \
" movd (%1), %%mm0 \n\t" \ " movd (%1), %%mm0 \n\t" \
" punpckl"#s" %%mm0, %%mm0 \n\t" \ " punpckldq %%mm0, %%mm0 \n\t" \
" movq %%mm0, (%0) \n\t" \ " movq %%mm0, (%0) \n\t" \
" add $4, %1 \n\t" \ " add $4, %1 \n\t" \
" add $8, %0 \n\t" " add $8, %0 \n\t"
#define MONO_TO_STEREO(s) \ #define HANDLE_SINGLE_wd() \
" mov %3, %2 \n\t" \ " movw (%1), %w3 \n\t" \
" sar $3, %2 \n\t" \ " movd %3, %%mm0 \n\t" \
" punpcklwd %%mm0, %%mm0 \n\t" \
" movd %%mm0, (%0) \n\t" \
" add $2, %1 \n\t" \
" add $4, %0 \n\t"
#define MONO_TO_STEREO(s,shift,mask) \
" mov %4, %2 \n\t" \
" sar $"#shift", %2 \n\t" \
" cmp $0, %2 \n\t" \ " cmp $0, %2 \n\t" \
" je 2f \n\t" \ " je 2f \n\t" \
"1: \n\t" \ "1: \n\t" \
@ -84,11 +92,11 @@
" dec %2 \n\t" \ " dec %2 \n\t" \
" jne 1b \n\t" \ " jne 1b \n\t" \
"2: \n\t" \ "2: \n\t" \
" mov %3, %2 \n\t" \ " mov %4, %2 \n\t" \
" and $7, %2 \n\t" \ " and $"#mask", %2 \n\t" \
" je 4f \n\t" \ " je 4f \n\t" \
"3: \n\t" \ "3: \n\t" \
HANDLE_SINGLE(s) \ HANDLE_SINGLE_##s() \
" dec %2 \n\t" \ " dec %2 \n\t" \
" jne 3b \n\t" \ " jne 3b \n\t" \
"4: \n\t" \ "4: \n\t" \
@ -96,14 +104,14 @@
#if defined (__i386__) || defined (__amd64__) #if defined (__i386__) || defined (__amd64__)
static void remap_mono_to_stereo_mmx (pa_remap_t *m, void *dst, const void *src, unsigned n) { static void remap_mono_to_stereo_mmx (pa_remap_t *m, void *dst, const void *src, unsigned n) {
pa_reg_x86 temp; pa_reg_x86 temp, temp2;
switch (*m->format) { switch (*m->format) {
case PA_SAMPLE_FLOAT32NE: case PA_SAMPLE_FLOAT32NE:
{ {
__asm__ __volatile__ ( __asm__ __volatile__ (
MONO_TO_STEREO(dq) /* do doubles to quads */ MONO_TO_STEREO(dq,3,7) /* do doubles to quads */
: "+r" (dst), "+r" (src), "=&r" (temp) : "+r" (dst), "+r" (src), "=&r" (temp), "=&r" (temp2)
: "r" ((pa_reg_x86)n) : "r" ((pa_reg_x86)n)
: "cc" : "cc"
); );
@ -112,8 +120,8 @@ static void remap_mono_to_stereo_mmx (pa_remap_t *m, void *dst, const void *src,
case PA_SAMPLE_S16NE: case PA_SAMPLE_S16NE:
{ {
__asm__ __volatile__ ( __asm__ __volatile__ (
MONO_TO_STEREO(wd) /* do words to doubles */ MONO_TO_STEREO(wd,4,15) /* do words to doubles */
: "+r" (dst), "+r" (src), "=&r" (temp) : "+r" (dst), "+r" (src), "=&r" (temp), "=&r" (temp2)
: "r" ((pa_reg_x86)n) : "r" ((pa_reg_x86)n)
: "cc" : "cc"
); );

34
src/pulsecore/remap_sse.c
View file

@ -65,16 +65,24 @@
" add $64, %1 \n\t" \ " add $64, %1 \n\t" \
" add $128, %0 \n\t" " add $128, %0 \n\t"
#define HANDLE_SINGLE(s) \ #define HANDLE_SINGLE_dq() \
" movd (%1), %%xmm0 \n\t" \ " movd (%1), %%xmm0 \n\t" \
" punpckl"#s" %%xmm0, %%xmm0 \n\t" \ " punpckldq %%xmm0, %%xmm0 \n\t" \
" movq %%xmm0, (%0) \n\t" \ " movq %%xmm0, (%0) \n\t" \
" add $4, %1 \n\t" \ " add $4, %1 \n\t" \
" add $8, %0 \n\t" " add $8, %0 \n\t"
#define MONO_TO_STEREO(s) \ #define HANDLE_SINGLE_wd() \
" mov %3, %2 \n\t" \ " movw (%1), %w3 \n\t" \
" sar $4, %2 \n\t" \ " movd %3, %%xmm0 \n\t" \
" punpcklwd %%xmm0, %%xmm0 \n\t" \
" movd %%xmm0, (%0) \n\t" \
" add $2, %1 \n\t" \
" add $4, %0 \n\t"
#define MONO_TO_STEREO(s,shift,mask) \
" mov %4, %2 \n\t" \
" sar $"#shift", %2 \n\t" \
" cmp $0, %2 \n\t" \ " cmp $0, %2 \n\t" \
" je 2f \n\t" \ " je 2f \n\t" \
"1: \n\t" \ "1: \n\t" \
@ -84,24 +92,24 @@
" dec %2 \n\t" \ " dec %2 \n\t" \
" jne 1b \n\t" \ " jne 1b \n\t" \
"2: \n\t" \ "2: \n\t" \
" mov %3, %2 \n\t" \ " mov %4, %2 \n\t" \
" and $15, %2 \n\t" \ " and $"#mask", %2 \n\t" \
" je 4f \n\t" \ " je 4f \n\t" \
"3: \n\t" \ "3: \n\t" \
HANDLE_SINGLE(s) \ HANDLE_SINGLE_##s() \
" dec %2 \n\t" \ " dec %2 \n\t" \
" jne 3b \n\t" \ " jne 3b \n\t" \
"4: \n\t" "4: \n\t"
static void remap_mono_to_stereo_sse (pa_remap_t *m, void *dst, const void *src, unsigned n) { static void remap_mono_to_stereo_sse (pa_remap_t *m, void *dst, const void *src, unsigned n) {
pa_reg_x86 temp; pa_reg_x86 temp, temp2;
switch (*m->format) { switch (*m->format) {
case PA_SAMPLE_FLOAT32NE: case PA_SAMPLE_FLOAT32NE:
{ {
__asm__ __volatile__ ( __asm__ __volatile__ (
MONO_TO_STEREO(dq) /* do doubles to quads */ MONO_TO_STEREO(dq,3,7) /* do doubles to quads */
: "+r" (dst), "+r" (src), "=&r" (temp) : "+r" (dst), "+r" (src), "=&r" (temp), "=&r" (temp2)
: "r" ((pa_reg_x86)n) : "r" ((pa_reg_x86)n)
: "cc" : "cc"
); );
@ -110,8 +118,8 @@ static void remap_mono_to_stereo_sse (pa_remap_t *m, void *dst, const void *src,
case PA_SAMPLE_S16NE: case PA_SAMPLE_S16NE:
{ {
__asm__ __volatile__ ( __asm__ __volatile__ (
MONO_TO_STEREO(wd) /* do words to doubles */ MONO_TO_STEREO(wd,4,15) /* do words to doubles */
: "+r" (dst), "+r" (src), "=&r" (temp) : "+r" (dst), "+r" (src), "=&r" (temp), "=&r" (temp2)
: "r" ((pa_reg_x86)n) : "r" ((pa_reg_x86)n)
: "cc" : "cc"
); );