remap: Change remapping function argument type from void to int16_t / float as approp...

[pulseaudio] / src / pulsecore / sconv_sse.c

diff --git a/src/pulsecore/sconv_sse.c b/src/pulsecore/sconv_sse.c
index 34b59fa54f86df0cfc63f1639cb2e61c2fb77e72..10943bad0bce2b38dc6d1ef93cc65b45220915a0 100644 (file)
--- a/src/pulsecore/sconv_sse.c
+++ b/src/pulsecore/sconv_sse.c
@@ -33,19 +33,15 @@
 #include "cpu-x86.h"
 #include "sconv.h"
 
-#if !defined(__APPLE__) && defined (__i386__) || defined (__amd64__)
+#if (!defined(__APPLE__) && !defined(__FreeBSD__) && defined (__i386__)) || defined (__amd64__)
 
-static const PA_DECLARE_ALIGNED (16, float, one[4]) = { 1.0, 1.0, 1.0, 1.0 };
-static const PA_DECLARE_ALIGNED (16, float, mone[4]) = { -1.0, -1.0, -1.0, -1.0 };
-static const PA_DECLARE_ALIGNED (16, float, scale[4]) = { 0x7fff, 0x7fff, 0x7fff, 0x7fff };
+static const PA_DECLARE_ALIGNED (16, float, scale[4]) = { 0x8000, 0x8000, 0x8000, 0x8000 };
 
 static void pa_sconv_s16le_from_f32ne_sse(unsigned n, const float *a, int16_t *b) {
     pa_reg_x86 temp, i;
 
     __asm__ __volatile__ (
         " movaps %5, %%xmm5             \n\t"
-        " movaps %6, %%xmm6             \n\t"
-        " movaps %7, %%xmm7             \n\t"
         " xor %0, %0                    \n\t"
 
         " mov %4, %1                    \n\t"
@@ -56,12 +52,8 @@ static void pa_sconv_s16le_from_f32ne_sse(unsigned n, const float *a, int16_t *b
         "1:                             \n\t"
         " movups (%q2, %0, 2), %%xmm0   \n\t" /* read 8 floats */
         " movups 16(%q2, %0, 2), %%xmm2 \n\t"
-        " minps  %%xmm5, %%xmm0         \n\t" /* clamp to 1.0 */
-        " minps  %%xmm5, %%xmm2         \n\t"
-        " maxps  %%xmm6, %%xmm0         \n\t" /* clamp to -1.0 */
-        " maxps  %%xmm6, %%xmm2         \n\t"
-        " mulps  %%xmm7, %%xmm0         \n\t" /* *= 0x7fff */
-        " mulps  %%xmm7, %%xmm2         \n\t"
+        " mulps  %%xmm5, %%xmm0         \n\t" /* *= 0x8000 */
+        " mulps  %%xmm5, %%xmm2         \n\t"
 
         " cvtps2pi %%xmm0, %%mm0        \n\t" /* low part to int */
         " cvtps2pi %%xmm2, %%mm2        \n\t"
@@ -73,7 +65,7 @@ static void pa_sconv_s16le_from_f32ne_sse(unsigned n, const float *a, int16_t *b
         " packssdw %%mm1, %%mm0         \n\t" /* pack parts */
         " packssdw %%mm3, %%mm2         \n\t"
         " movq     %%mm0, (%q3, %0)     \n\t"
-        " movq    %%mm2, 8(%q3, %0)     \n\t"
+        " movq     %%mm2, 8(%q3, %0)    \n\t"
 
         " add $16, %0                   \n\t"
         " dec %1                        \n\t"
@@ -82,24 +74,30 @@ static void pa_sconv_s16le_from_f32ne_sse(unsigned n, const float *a, int16_t *b
         "2:                             \n\t"
         " mov %4, %1                    \n\t" /* prepare for leftovers */
         " and $7, %1                    \n\t"
-        " je 4f                         \n\t"
+        " je 5f                         \n\t"
 
         "3:                             \n\t"
         " movss (%q2, %0, 2), %%xmm0    \n\t"
-        " minss  %%xmm5, %%xmm0         \n\t"
-        " maxss  %%xmm6, %%xmm0         \n\t"
-        " mulss  %%xmm7, %%xmm0         \n\t"
+        " mulss  %%xmm5, %%xmm0         \n\t"
+        " cvtss2si %%xmm0, %4           \n\t"
+        " add $0x8000, %4               \n\t" /* check for saturation */
+        " and $~0xffff, %4              \n\t"
         " cvtss2si %%xmm0, %4           \n\t"
-        " movw  %w4, (%q3, %0)          \n\t"
+        " je 4f                         \n\t"
+        " sar $31, %4                   \n\t"
+        " xor $0x7fff, %4               \n\t"
+
+        "4:                             \n\t"
+        " movw  %w4, (%q3, %0)          \n\t" /* store leftover */
         " add $2, %0                    \n\t"
         " dec %1                        \n\t"
         " jne 3b                        \n\t"
 
-        "4:                             \n\t"
+        "5:                             \n\t"
         " emms                          \n\t"
 
         : "=&r" (i), "=&r" (temp)
-        : "r" (a), "r" (b), "r" ((pa_reg_x86)n), "m" (*one), "m" (*mone), "m" (*scale)
+        : "r" (a), "r" (b), "r" ((pa_reg_x86)n), "m" (*scale)
         : "cc", "memory"
     );
 }
@@ -109,8 +107,6 @@ static void pa_sconv_s16le_from_f32ne_sse2(unsigned n, const float *a, int16_t *
 
     __asm__ __volatile__ (
         " movaps %5, %%xmm5             \n\t"
-        " movaps %6, %%xmm6             \n\t"
-        " movaps %7, %%xmm7             \n\t"
         " xor %0, %0                    \n\t"
 
         " mov %4, %1                    \n\t"
@@ -121,12 +117,8 @@ static void pa_sconv_s16le_from_f32ne_sse2(unsigned n, const float *a, int16_t *
         "1:                             \n\t"
         " movups (%q2, %0, 2), %%xmm0   \n\t" /* read 8 floats */
         " movups 16(%q2, %0, 2), %%xmm2 \n\t"
-        " minps  %%xmm5, %%xmm0         \n\t" /* clamp to 1.0 */
-        " minps  %%xmm5, %%xmm2         \n\t"
-        " maxps  %%xmm6, %%xmm0         \n\t" /* clamp to -1.0 */
-        " maxps  %%xmm6, %%xmm2         \n\t"
-        " mulps  %%xmm7, %%xmm0         \n\t" /* *= 0x7fff */
-        " mulps  %%xmm7, %%xmm2         \n\t"
+        " mulps  %%xmm5, %%xmm0         \n\t" /* *= 0x8000 */
+        " mulps  %%xmm5, %%xmm2         \n\t"
 
         " cvtps2dq %%xmm0, %%xmm0       \n\t"
         " cvtps2dq %%xmm2, %%xmm2       \n\t"
@@ -141,23 +133,29 @@ static void pa_sconv_s16le_from_f32ne_sse2(unsigned n, const float *a, int16_t *
         "2:                             \n\t"
         " mov %4, %1                    \n\t" /* prepare for leftovers */
         " and $7, %1                    \n\t"
-        " je 4f                         \n\t"
+        " je 5f                         \n\t"
 
         "3:                             \n\t"
         " movss (%q2, %0, 2), %%xmm0    \n\t"
-        " minss  %%xmm5, %%xmm0         \n\t"
-        " maxss  %%xmm6, %%xmm0         \n\t"
-        " mulss  %%xmm7, %%xmm0         \n\t"
+        " mulss  %%xmm5, %%xmm0         \n\t"
+        " cvtss2si %%xmm0, %4           \n\t"
+        " add $0x8000, %4               \n\t"
+        " and $~0xffff, %4              \n\t" /* check for saturation */
         " cvtss2si %%xmm0, %4           \n\t"
-        " movw  %w4, (%q3, %0)          \n\t"
+        " je 4f                         \n\t"
+        " sar $31, %4                   \n\t"
+        " xor $0x7fff, %4               \n\t"
+
+        "4:                             \n\t"
+        " movw  %w4, (%q3, %0)          \n\t" /* store leftover */
         " add $2, %0                    \n\t"
         " dec %1                        \n\t"
         " jne 3b                        \n\t"
 
-        "4:                             \n\t"
+        "5:                             \n\t"
 
         : "=&r" (i), "=&r" (temp)
-        : "r" (a), "r" (b), "r" ((pa_reg_x86)n), "m" (*one), "m" (*mone), "m" (*scale)
+        : "r" (a), "r" (b), "r" ((pa_reg_x86)n), "m" (*scale)
         : "cc", "memory"
     );
 }
@@ -165,7 +163,7 @@ static void pa_sconv_s16le_from_f32ne_sse2(unsigned n, const float *a, int16_t *
 #endif /* defined (__i386__) || defined (__amd64__) */
 
 void pa_convert_func_init_sse(pa_cpu_x86_flag_t flags) {
-#if !defined(__APPLE__) && defined (__i386__) || defined (__amd64__)
+#if (!defined(__APPLE__) && !defined(__FreeBSD__) && defined (__i386__)) || defined (__amd64__)
 
     if (flags & PA_CPU_X86_SSE2) {
         pa_log_info("Initialising SSE2 optimized conversions.");