From: Mark Harmstone <hellas@burntcomma.com>
Subject: [PATCH] dsound: Apply filters before sound is multiplied to speakers. (try 2)
Message-Id: <550B47B7.5090203@burntcomma.com>
Date: Thu, 19 Mar 2015 22:03:35 +0000

Now only copies to a temporary buffer if we have any filters enabled.

---
 dlls/dsound/mixer.c | 97 ++++++++++++++++++++++++++++++++++++++++++-----------
 1 file changed, 78 insertions(+), 19 deletions(-)

diff --git a/dlls/dsound/mixer.c b/dlls/dsound/mixer.c
index 8dbd15a..09155e1 100644
--- a/dlls/dsound/mixer.c
+++ b/dlls/dsound/mixer.c
@@ -266,15 +266,62 @@ static inline float get_current_sample(const IDirectSoundBufferImpl *dsb,
     return dsb->get(dsb, mixpos % dsb->buflen, channel);
 }
 
+static void apply_filters(IDirectSoundBufferImpl *dsb, float* buf, UINT count)
+{
+    int i;
+    HRESULT hr;
+
+    if (count > 0) {
+        for (i = 0; i < dsb->num_filters; i++) {
+            if (dsb->filters[i].inplace) {
+                hr = IMediaObjectInPlace_Process(dsb->filters[i].inplace, count * dsb->mix_channels * sizeof(float),
+                                                (BYTE*)buf, 0, DMO_INPLACE_NORMAL);
+
+                if (FAILED(hr))
+                    WARN("IMediaObjectInPlace_Process failed for filter %u\n", i);
+            } else
+                WARN("filter %u has no inplace object - unsupported\n", i);
+        }
+    }
+}
+
 static UINT cp_fields_noresample(IDirectSoundBufferImpl *dsb, UINT count)
 {
     UINT istride = dsb->pwfx->nBlockAlign;
     UINT ostride = dsb->device->pwfx->nChannels * sizeof(float);
     DWORD channel, i;
-    for (i = 0; i < count; i++)
-        for (channel = 0; channel < dsb->mix_channels; channel++)
-            dsb->put(dsb, i * ostride, channel, get_current_sample(dsb,
-                    dsb->sec_mixpos + i * istride, channel));
+    BOOL using_filters = dsb->num_filters > 0;
+    float *buf, *buftmp;
+
+    if (using_filters) {
+        buf = HeapAlloc(GetProcessHeap(), 0, count * dsb->mix_channels * sizeof(float));
+        buftmp = buf;
+
+        for (i = 0; i < count; i++) {
+            for (channel = 0; channel < dsb->mix_channels; channel++) {
+                *buftmp = get_current_sample(dsb, dsb->sec_mixpos + i*istride, channel);
+                buftmp++;
+            }
+        }
+
+        apply_filters(dsb, buf, count);
+
+        buftmp = buf;
+        for (i = 0; i < count; i++) {
+            for (channel = 0; channel < dsb->mix_channels; channel++) {
+                dsb->put(dsb, i * ostride, channel, *buftmp);
+                buftmp++;
+            }
+        }
+
+        HeapFree(GetProcessHeap(), 0, buf);
+    } else {
+        for (i = 0; i < count; i++)
+            for (channel = 0; channel < dsb->mix_channels; channel++)
+                dsb->put(dsb, i * ostride, channel, get_current_sample(dsb,
+                        dsb->sec_mixpos + i * istride, channel));
+    }
+
     return count;
 }
 
@@ -299,6 +346,9 @@ static UINT cp_fields_resample(IDirectSoundBufferImpl *dsb, UINT count, LONG64 *
     float* fir_copy = HeapAlloc(GetProcessHeap(), 0,
             sizeof(float) * fir_cachesize);
 
+    BOOL using_filters = dsb->num_filters > 0;
+    float *buf, *buftmp;
+
     /* Important: this buffer MUST be non-interleaved
      * if you want -msse3 to have any effect.
      * This is good for CPU cache effects, too.
@@ -309,6 +359,11 @@ static UINT cp_fields_resample(IDirectSoundBufferImpl *dsb, UINT count, LONG64 *
             *(itmp++) = get_current_sample(dsb,
                     dsb->sec_mixpos + i * istride, channel);
 
+    if (using_filters) {
+        buf = HeapAlloc(GetProcessHeap(), 0, sizeof(float) * count * dsb->mix_channels);
+        buftmp = buf;
+    }
+
     for(i = 0; i < count; ++i) {
         UINT int_fir_steps = (freqAcc_start + i * dsb->freqAdjustNum) * dsbfirstep / dsb->freqAdjustDen;
         float total_fir_steps = (freqAcc_start + i * dsb->freqAdjustNum) * dsbfirstep / (float)dsb->freqAdjustDen;
@@ -332,10 +387,28 @@ static UINT cp_fields_resample(IDirectSoundBufferImpl *dsb, UINT count, LONG64 *
             float* cache = &intermediate[channel * required_input + ipos];
             for (j = 0; j < fir_used; j++)
                 sum += fir_copy[j] * cache[j];
-            dsb->put(dsb, i * ostride, channel, sum * dsb->firgain);
+
+            if (using_filters)
+                *(buftmp++) = sum * dsb->firgain;
+            else
+                dsb->put(dsb, i * ostride, channel, sum * dsb->firgain);
         }
     }
 
+    if (using_filters) {
+        apply_filters(dsb, buf, count);
+
+        buftmp = buf;
+        for (i = 0; i < count; i++) {
+            for (channel = 0; channel < dsb->mix_channels; channel++) {
+                dsb->put(dsb, i * ostride, channel, *buftmp);
+                buftmp++;
+            }
+        }
+
+        HeapFree(GetProcessHeap(), 0, buf);
+    }
+
     *freqAccNum = freqAcc_end % dsb->freqAdjustDen;
 
     HeapFree(GetProcessHeap(), 0, fir_copy);
@@ -397,8 +470,6 @@ static inline DWORD DSOUND_BufPtrDiff(DWORD buflen, DWORD ptr1, DWORD ptr2)
 static void DSOUND_MixToTemporary(IDirectSoundBufferImpl *dsb, DWORD frames)
 {
 	UINT size_bytes = frames * sizeof(float) * dsb->device->pwfx->nChannels;
-	HRESULT hr;
-	int i;
 
 	if (dsb->device->tmp_buffer_len < size_bytes || !dsb->device->tmp_buffer)
 	{
@@ -410,18 +481,6 @@ static void DSOUND_MixToTemporary(IDirectSoundBufferImpl *dsb, DWORD frames)
 	}
 
 	cp_fields(dsb, frames, &dsb->freqAccNum);
-
-	if (size_bytes > 0) {
-		for (i = 0; i < dsb->num_filters; i++) {
-			if (dsb->filters[i].inplace) {
-				hr = IMediaObjectInPlace_Process(dsb->filters[i].inplace, size_bytes, (BYTE*)dsb->device->tmp_buffer, 0, DMO_INPLACE_NORMAL);
-
-				if (FAILED(hr))
-					WARN("IMediaObjectInPlace_Process failed for filter %u\n", i);
-			} else
-				WARN("filter %u has no inplace object - unsupported\n", i);
-		}
-	}
 }
 
 static void DSOUND_MixerVol(const IDirectSoundBufferImpl *dsb, INT frames)