summaryrefslogtreecommitdiffstats
path: root/src/pulsecore/remap_mmx.c
diff options
context:
space:
mode:
Diffstat (limited to 'src/pulsecore/remap_mmx.c')
-rw-r--r--src/pulsecore/remap_mmx.c65
1 files changed, 36 insertions, 29 deletions
diff --git a/src/pulsecore/remap_mmx.c b/src/pulsecore/remap_mmx.c
index 00252dac..79e4f1fc 100644
--- a/src/pulsecore/remap_mmx.c
+++ b/src/pulsecore/remap_mmx.c
@@ -51,7 +51,7 @@
" punpckl"#s" %%mm4, %%mm4 \n\t" \
" punpckh"#s" %%mm5, %%mm5 \n\t" \
" punpckl"#s" %%mm6, %%mm6 \n\t" \
- " punpckh"#s" %%mm7, %%mm7 \n\t" \
+ " punpckh"#s" %%mm7, %%mm7 \n\t"
#define STORE_SAMPLES \
" movq %%mm0, (%0) \n\t" \
@@ -65,46 +65,53 @@
" add $32, %1 \n\t" \
" add $64, %0 \n\t"
-#define HANDLE_SINGLE(s) \
+#define HANDLE_SINGLE_dq() \
" movd (%1), %%mm0 \n\t" \
- " movq %%mm0, %%mm1 \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" \
- " cmp $0, %2 \n\t" \
- " je 2f \n\t" \
- "1: \n\t" \
- LOAD_SAMPLES \
- UNPACK_SAMPLES(s) \
- STORE_SAMPLES \
- " dec %2 \n\t" \
- " jne 1b \n\t" \
- "2: \n\t" \
- " mov %3, %2 \n\t" \
- " and $7, %2 \n\t" \
- " je 4f \n\t" \
- "3: \n\t" \
- HANDLE_SINGLE(s) \
- " dec %2 \n\t" \
- " jne 3b \n\t" \
- "4: \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" \
+ LOAD_SAMPLES \
+ UNPACK_SAMPLES(s) \
+ STORE_SAMPLES \
+ " dec %2 \n\t" \
+ " jne 1b \n\t" \
+ "2: \n\t" \
+ " mov %4, %2 \n\t" \
+ " and $"#mask", %2 \n\t" \
+ " je 4f \n\t" \
+ "3: \n\t" \
+ HANDLE_SINGLE_##s() \
+ " dec %2 \n\t" \
+ " jne 3b \n\t" \
+ "4: \n\t" \
" emms \n\t"
#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"
);
@@ -113,8 +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"
);