Changeset 887

Timestamp:

02/24/08 07:40:16 (13 years ago)

Author:

ppalmers

Message:

implement clipping for float data

Files:

• trunk/libffado/config.h.in (modified) (2 diffs)
• trunk/libffado/SConstruct (modified) (1 diff)
• trunk/libffado/src/libstreaming/amdtp/AmdtpTransmitStreamProcessor.cpp (modified) (5 diffs)

trunk/libffado/config.h.in

@@ -26 +26 @@
-/* Define to the api version */
-#define FFADO_API_VERSION $FFADO_API_VERSION
-
-/* Define to 1 if SSE assembly is available. */
-// broken
-//#define USE_SSE $USE_SSE
-#define USE_SSE 0
-//#define USE_SSE2 $USE_SSE2
-#define USE_SSE2 0
-
-#define CACHEDIR "~/.ffado"
@@ -115 +108 @@
-#define AMDTP_MAX_CYCLES_TO_TRANSMIT_EARLY                  6
-
+// ensure that the AMDTP SP clips all float values to [-1.0..1.0]
+#define AMDTP_CLIP_FLOATS                                   1
+
+
-#define MOTU_TRANSMIT_TRANSFER_DELAY    (11776U)
-

trunk/libffado/SConstruct

@@ -24 +24 @@
-
-FFADO_API_VERSION="7"
-5
+6
-
-import os

trunk/libffado/src/libstreaming/amdtp/AmdtpTransmitStreamProcessor.cpp

@@ -529 +529 @@
-    const __m128 mult = _mm_set_ps(AMDTP_FLOAT_MULTIPLIER, AMDTP_FLOAT_MULTIPLIER, AMDTP_FLOAT_MULTIPLIER, AMDTP_FLOAT_MULTIPLIER);
-
+#if AMDTP_CLIP_FLOATS
+    const __m128 v_max = _mm_set_ps(1.0, 1.0, 1.0, 1.0);
+    const __m128 v_min = _mm_set_ps(-1.0, -1.0, -1.0, -1.0);
+#endif
+
-    // this assumes that audio ports are sorted by position,
-    // and that there are no gaps
@@ -563 +568 @@
-            __m128i *target = (__m128i*)target_event;
-            __m128i v_int;
+
+            // clip
+#if AMDTP_CLIP_FLOATS
+            // implement sample<min?min:sample
+            // and sample>max?max:sample
+            // we use separate masks since that allows the
+            // compiler/cpu to do more out-of-order optimization
+
+            // is any of the pieces less than the minimum?
+            // or larger than the maximum?
+            __m128 mask1 = _mm_cmplt_ps(v_float, v_min);
+            __m128 mask2 = _mm_cmpgt_ps(v_float, v_max);
+            // clip the values that need to be clipped
+            // pass the values that don't
+            v_float = _mm_or_ps(_mm_andnot_ps(mask1, v_float), _mm_and_ps(mask1, v_min));
+            v_float = _mm_or_ps(_mm_andnot_ps(mask2, v_float), _mm_and_ps(mask2, v_max));
+#endif
-
-            // multiply
@@ -616 +638 @@
-                tmp_values[3] = *buffer;
-                buffer++;
-
+
-                // now do the SSE based conversion/labeling
-                __m128 v_float = *((__m128*)tmp_values);
-                __m128i v_int;
-
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
-                // multiply
-                v_float = _mm_mul_ps(v_float, mult);
@@ -653 +691 @@
-            for(;j < nevents; j += 1) {
-                float *in = (float *)buffer;
+#if AMDTP_CLIP_FLOATS
+                if(*in > 1.0) *in=1.0;
+                if(*in < -1.0) *in=-1.0;
+#endif
-                float v = (*in) * AMDTP_FLOAT_MULTIPLIER;
-                unsigned int tmp = ((int) v);
@@ -898 +940 @@
-            {
-                float *in = (float *)buffer;
+#if AMDTP_CLIP_FLOATS
+                if(*in > 1.0) *in=1.0;
+                if(*in < -1.0) *in=-1.0;
+#endif
-                float v = (*in) * AMDTP_FLOAT_MULTIPLIER;
-                unsigned int tmp = ((int) v);