Changeset 777

Timestamp:

12/26/07 13:59:21 (16 years ago)

Author:

ppalmers

Message:

port freebob SSE optimized event encoding functions (untested)

Files:

• trunk/libffado/SConstruct (modified) (1 diff)
• trunk/libffado/src/debugmodule/debugmodule.h (modified) (1 diff)
• trunk/libffado/src/devicemanager.cpp (modified) (13 diffs)
• trunk/libffado/src/devicemanager.h (modified) (1 diff)
• trunk/libffado/src/libieee1394/ieee1394service.cpp (modified) (3 diffs)
• trunk/libffado/src/libieee1394/IsoHandler.cpp (modified) (1 diff)
• trunk/libffado/src/libstreaming/amdtp/AmdtpReceiveStreamProcessor.cpp (modified) (5 diffs)
• trunk/libffado/src/libstreaming/amdtp/AmdtpTransmitStreamProcessor.cpp (modified) (4 diffs)

trunk/libffado/SConstruct

@@ -285 +285 @@
-                opt_flags.append ("-march=i686")
-    
-
+
+
-        opt_flags.extend (["-msse", "-mfpmath=sse"])
-        env['USE_SSE'] = 1

trunk/libffado/src/debugmodule/debugmodule.h

@@ -53 +53 @@
-#define MB_BUFFERSIZE       DEBUG_MAX_MESSAGE_LENGTH
-
-
+
+
+
+
-#ifdef IMPLEMENT_BACKLOG
-// the backlog is a similar buffer as the message buffer

trunk/libffado/src/devicemanager.cpp

@@ -81 +81 @@
-DeviceManager::DeviceManager()
-    : Control::Container("devicemanager")
+    , m_processorManager( new Streaming::StreamProcessorManager() )
-{
-    addOption(Util::OptionContainer::Option("slaveMode",false));
@@ -88 +89 @@
-DeviceManager::~DeviceManager()
-{
+    delete m_processorManager;
+
-    for ( FFADODeviceVectorIterator it = m_avDevices.begin();
-          it != m_avDevices.end();
@@ -98 +101 @@
-    }
-
+    for ( FunctorVectorIterator it = m_busreset_functors.begin();
+          it != m_busreset_functors.end();
+          ++it )
+    {
+        delete *it;
+    }
+
-    for ( Ieee1394ServiceVectorIterator it = m_1394Services.begin();
-          it != m_1394Services.end();
@@ -104 +114 @@
-        delete *it;
-    }
-    for ( FunctorVectorIterator it = m_busreset_functors.begin();
-          it != m_busreset_functors.end();
-          ++it )
-    {
-        delete *it;
-    }
-}
-
-bool
-DeviceManager::setThreadParameters(bool rt, int priority) {
-.
+->
-        debugError("Could not set processor manager thread parameters\n");
-        return false;
@@ -419 +423 @@
-
-        debugOutput(DEBUG_LEVEL_VERBOSE, "Setting samplerate to %d for (%p)\n",
-.
+->
-
-        // Set the device's sampling rate to that requested
-        // FIXME: does this really belong here?  If so we need to handle errors.
-.
+->
-            debugOutput(DEBUG_LEVEL_VERBOSE, " => Retry setting samplerate to %d for (%p)\n",
-.
+->
-
-            // try again:
-.
-.
+->
+->
-                return false;
-            }
@@ -438 +442 @@
-
-    // set the sync source
-.
+->
-        debugWarning("Could not set processorManager sync source (%p)\n",
-            getSyncSource());
@@ -448 +452 @@
-DeviceManager::prepareStreaming()
-{
-.
+->
-        debugFatal("Could not prepare streaming...\n");
-        return false;
@@ -500 +504 @@
-    }
-
-.
+->
-        return true;
-    } else {
@@ -517 +521 @@
-{
-    bool result = true;
-.
+->
-
-    // create the connections for all devices
@@ -550 +554 @@
-bool
-DeviceManager::waitForPeriod() {
-.
+->
-        return true;
-    } else {
-        debugWarning("XRUN detected\n");
-        // do xrun recovery
-.
+->
-        return false;
-    }
@@ -562 +566 @@
-bool
-DeviceManager::setStreamingParams(unsigned int period, unsigned int rate, unsigned int nb_buffers) {
-.
-.
-.
+->
+->
+->
-    return true;
-}
@@ -756 +760 @@
-    setDebugLevel(l);
-    Control::Element::setVerboseLevel(l);
-.
+->
-    for ( FFADODeviceVectorIterator it = m_avDevices.begin();
-          it != m_avDevices.end();
@@ -810 +814 @@
-void
-DeviceManager::showStreamingInfo() {
-.
-
+->
+

trunk/libffado/src/devicemanager.h

@@ -117 +117 @@
-public: // FIXME: this should be better
-    Streaming::StreamProcessorManager&  getStreamProcessorManager() 
-
+*
-private:
-
+*
-protected:
-    std::vector<std::string>          m_SpecStrings;

trunk/libffado/src/libieee1394/ieee1394service.cpp

@@ -102 +102 @@
-    delete m_pCTRHelper;
-    delete m_pIsoManager;
-    delete m_pTimeSource;
-    stopRHThread();
-    for ( arm_handler_vec_t::iterator it = m_armHandlers.begin();
@@ -116 +115 @@
-    }
-
+    delete m_pTimeSource;
-    if ( m_handle ) {
-        raw1394_destroy_handle( m_handle );
@@ -352 +352 @@
-uint64_t
-Ieee1394Service::getCurrentTimeAsUsecs() {
-
-
+
+
+
+
+
+
-}
-

trunk/libffado/src/libieee1394/IsoHandler.cpp

@@ -217 +217 @@
-                poll_exit-poll_enter, iter_exit-iter_enter);
-#else
-itself blocks if nothing
+blocks if no 1394 data
-    // so poll'ing is not really necessary
-    bool result = iterate();

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

@@ -21 +21 @@
- *
- */
+#include "config.h"
-
-#include "AmdtpReceiveStreamProcessor.h"
@@ -250 +251 @@
-}
-
+#ifdef USE_SSE
+typedef float v4sf __attribute__ ((vector_size (16)));
+typedef int v4si __attribute__ ((vector_size (16)));
+typedef int v2si __attribute__ ((vector_size (8)));
+
-int
-AmdtpReceiveStreamProcessor::decodeMBLAEventsToPort(
@@ -260 +266 @@
-    target_event=(quadlet_t *)(data + p->getPosition());
-
+    static const float multiplier = 1.0f / (float)(0x7FFFFF);
+    static const float sse_multiplier[4] __attribute__((aligned(16))) = {
+            1.0f / (float)(0x7FFFFF),
+            1.0f / (float)(0x7FFFFF),
+            1.0f / (float)(0x7FFFFF),
+            1.0f / (float)(0x7FFFFF)
+    };
+    unsigned int tmp[4];
+
-    switch(p->getDataType()) {
-        default:
@@ -279 +294 @@
-        case Port::E_Float:
-            {
-                const float multiplier = 1.0f / (float)(0x7FFFFF);
-                float *buffer=(float *)(p->getBufferAddress());
-
-                assert(nevents + offset <= p->getBufferSize());
-
-
-
-
-
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
-                    unsigned int v = ntohl(*target_event) & 0x00FFFFFF;
-                    // sign-extend highest bit of 24-bit int
-                    int tmp = (int)(v << 8) / 256;
-
-                    *buffer = tmp * multiplier;
-
+                    buffer++;
+                    target_event += m_dimension;
+                }
+                asm volatile("emms");
+                break;
+            }
+            break;
+    }
+
+    return 0;
+}
+
+#else
+
+int
+AmdtpReceiveStreamProcessor::decodeMBLAEventsToPort(
+                       AmdtpAudioPort *p, quadlet_t *data,
+                       unsigned int offset, unsigned int nevents)
+{
+    unsigned int j=0;
+    quadlet_t *target_event;
+
+    target_event=(quadlet_t *)(data + p->getPosition());
+
+    switch(p->getDataType()) {
+        default:
+        case Port::E_Int24:
+            {
+                quadlet_t *buffer=(quadlet_t *)(p->getBufferAddress());
+
+                assert(nevents + offset <= p->getBufferSize());
+
+                buffer+=offset;
+
+                for(j = 0; j < nevents; j += 1) { // decode max nsamples
+                    *(buffer)=(ntohl((*target_event) ) & 0x00FFFFFF);
-                    buffer++;
-                    target_event+=m_dimension;
@@ -299 +374 @@
-            }
-            break;
+        case Port::E_Float:
+            {
+                const float multiplier = 1.0f / (float)(0x7FFFFF);
+                float *buffer=(float *)(p->getBufferAddress());
+
+                assert(nevents + offset <= p->getBufferSize());
+
+                buffer+=offset;
+
+                for(j = 0; j < nevents; j += 1) { // decode max nsamples
+
+                    unsigned int v = ntohl(*target_event) & 0x00FFFFFF;
+                    // sign-extend highest bit of 24-bit int
+                    int tmp = (int)(v << 8) / 256;
+
+                    *buffer = tmp * multiplier;
+
+                    buffer++;
+                    target_event+=m_dimension;
+                }
+            }
+            break;
-    }
-
-    return 0;
-}
+#endif
-} // end of namespace Streaming

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

@@ -22 +22 @@
- */
-
+#include "config.h"
-#include "AmdtpTransmitStreamProcessor.h"
-#include "AmdtpPort.h"
@@ -99 +100 @@
-    const int max_cycles_to_transmit_early = 2;
-
-try_block_of_frames:
-    debugOutput ( DEBUG_LEVEL_ULTRA_VERBOSE, "Try for cycle %d\n", cycle );
-    // check whether the packet buffer has packets for us to send.
@@ -591 +591 @@
-}
-
+#ifdef USE_SSE
+typedef float v4sf __attribute__ ((vector_size (16)));
+typedef int v4si __attribute__ ((vector_size (16)));
+typedef int v2si __attribute__ ((vector_size (8)));
+
+int AmdtpTransmitStreamProcessor::encodePortToMBLAEvents ( AmdtpAudioPort *p, quadlet_t *data,
+        unsigned int offset, unsigned int nevents )
+{
+    static const float sse_multiplier[4] __attribute__((aligned(16))) = {
+        (float)(0x7FFFFF00),
+        (float)(0x7FFFFF00),
+        (float)(0x7FFFFF00),
+        (float)(0x7FFFFF00)
+    };
+
+    static const int sse_mask[4] __attribute__((aligned(16))) = {
+        0x40000000,  0x40000000,  0x40000000,  0x40000000
+    };
+
+    unsigned int out[4];
+
+    unsigned int j=0;
+    unsigned int read=0;
+
+    quadlet_t *target_event;
+
+    target_event= ( quadlet_t * ) ( data + p->getPosition() );
+
+    switch ( p->getDataType() )
+    {
+        default:
+        case Port::E_Int24:
+        {
+            quadlet_t *buffer= ( quadlet_t * ) ( p->getBufferAddress() );
+
+            assert ( nevents + offset <= p->getBufferSize() );
+
+            buffer+=offset;
+
+            for ( j = 0; j < nevents; j += 1 )   // decode max nsamples
+            {
+                *target_event = htonl ( ( * ( buffer ) & 0x00FFFFFF ) | 0x40000000 );
+                buffer++;
+                target_event += m_dimension;
+            }
+        }
+        break;
+        case Port::E_Float:
+        {
+            const float multiplier = ( float ) ( 0x7FFFFF00 );
+            float *buffer= ( float * ) ( p->getBufferAddress() );
+
+            assert ( nevents + offset <= p->getBufferSize() );
+
+            buffer+=offset;
+
+            j=0;
+            if(read>3) {
+                for (j = 0; j < read-3; j += 4) {
+                    asm("movups %[floatbuff], %%xmm0\n\t"
+                            "mulps %[ssemult], %%xmm0\n\t"
+                            "cvttps2pi %%xmm0, %[out1]\n\t"
+                            "movhlps %%xmm0, %%xmm0\n\t"
+                            "psrld $8, %[out1]\n\t"
+                            "cvttps2pi %%xmm0, %[out2]\n\t"
+                            "por %[mmxmask], %[out1]\n\t"
+                            "psrld $8, %[out2]\n\t"
+                            "por %[mmxmask], %[out2]\n\t"
+                        : [out1] "=&y" (*(v2si*)&out[0]),
+                    [out2] "=&y" (*(v2si*)&out[2])
+                        : [floatbuff] "m" (*(v4sf*)buffer),
+                    [ssemult] "x" (*(v4sf*)sse_multiplier),
+                    [mmxmask] "y" (*(v2si*)sse_mask)
+                        : "xmm0");
+                    buffer += 4;
+                    *target_event = htonl(out[0]);
+                    target_event += m_dimension;
+                    *target_event = htonl(out[1]);
+                    target_event += m_dimension;
+                    *target_event = htonl(out[2]);
+                    target_event += m_dimension;
+                    *target_event = htonl(out[3]);
+                    target_event += m_dimension;
+                }
+            }
+            for(; j < read; ++j) {
+            // don't care for overflow
+                float v = *buffer * multiplier;  // v: -231 .. 231
+                unsigned int tmp = (int)v;
+                *target_event = htonl((tmp >> 8) | 0x40000000);
+    
+                buffer++;
+                target_event += m_dimension;
+            }
+
+            asm volatile("emms");
+            break;
+        }
+        break;
+    }
+
+    return 0;
+}
+
+#else
-
-int AmdtpTransmitStreamProcessor::encodePortToMBLAEvents ( AmdtpAudioPort *p, quadlet_t *data,
@@ -646 +751 @@
-    return 0;
-}
+#endif
+
-int AmdtpTransmitStreamProcessor::encodeSilencePortToMBLAEvents ( AmdtpAudioPort *p, quadlet_t *data,
-        unsigned int offset, unsigned int nevents )