diff options
| author | Wim Taymans <wim.taymans@collabora.co.uk> | 2009-08-29 12:22:42 +0200 |
|---|---|---|
| committer | Lennart Poettering <lennart@poettering.net> | 2009-08-31 21:50:56 +0200 |
| commit | dc221f204b89fca85c0125e55f3afea4a807ffa7 (patch) | |
| tree | c265009f484b32562a7e7729d33ffe6e06e657f3 /src/pulsecore/remap_mmx.c | |
| parent | dee2aa3f0564caed698e600963b592d50cda068c (diff) | |
remap: fix counters for mmx and sse remap
Take the size of the sample into account when calculating the amount of samples
we process in parallel.
Diffstat (limited to 'src/pulsecore/remap_mmx.c')
| -rw-r--r-- | src/pulsecore/remap_mmx.c | 34 |
1 files changed, 21 insertions, 13 deletions
diff --git a/src/pulsecore/remap_mmx.c b/src/pulsecore/remap_mmx.c index b5fe82ee7..79e4f1fc2 100644 --- a/src/pulsecore/remap_mmx.c +++ b/src/pulsecore/remap_mmx.c @@ -65,16 +65,24 @@ " add $32, %1 \n\t" \ " add $64, %0 \n\t" -#define HANDLE_SINGLE(s) \ +#define HANDLE_SINGLE_dq() \ " movd (%1), %%mm0 \n\t" \ - " punpckl"#s" %%mm0, %%mm0 \n\t" \ + " punpckldq %%mm0, %%mm0 \n\t" \ " movq %%mm0, (%0) \n\t" \ " add $4, %1 \n\t" \ " add $8, %0 \n\t" -#define MONO_TO_STEREO(s) \ - " mov %3, %2 \n\t" \ - " sar $3, %2 \n\t" \ +#define HANDLE_SINGLE_wd() \ + " movw (%1), %w3 \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" \ " je 2f \n\t" \ "1: \n\t" \ @@ -84,11 +92,11 @@ " dec %2 \n\t" \ " jne 1b \n\t" \ "2: \n\t" \ - " mov %3, %2 \n\t" \ - " and $7, %2 \n\t" \ + " mov %4, %2 \n\t" \ + " and $"#mask", %2 \n\t" \ " je 4f \n\t" \ "3: \n\t" \ - HANDLE_SINGLE(s) \ + HANDLE_SINGLE_##s() \ " dec %2 \n\t" \ " jne 3b \n\t" \ "4: \n\t" \ @@ -96,14 +104,14 @@ #if defined (__i386__) || defined (__amd64__) 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) { case PA_SAMPLE_FLOAT32NE: { __asm__ __volatile__ ( - MONO_TO_STEREO(dq) /* do doubles to quads */ - : "+r" (dst), "+r" (src), "=&r" (temp) + MONO_TO_STEREO(dq,3,7) /* do doubles to quads */ + : "+r" (dst), "+r" (src), "=&r" (temp), "=&r" (temp2) : "r" ((pa_reg_x86)n) : "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: { __asm__ __volatile__ ( - MONO_TO_STEREO(wd) /* do words to doubles */ - : "+r" (dst), "+r" (src), "=&r" (temp) + MONO_TO_STEREO(wd,4,15) /* do words to doubles */ + : "+r" (dst), "+r" (src), "=&r" (temp), "=&r" (temp2) : "r" ((pa_reg_x86)n) : "cc" ); |