2.0.0: Fixed potential audio-mixer bug which could happen when parallel loop was...

git-svn-id: https://casparcg.svn.sourceforge.net/svnroot/casparcg/server/branches/2.0.0.2@1381 362d55ac-95cf-4e76-9f9a-cbaa9c17b72d

diff --git a/core/mixer/audio/audio_mixer.cpp b/core/mixer/audio/audio_mixer.cpp
index 4dccf01a7a5b652e177184b1b3ccbdbc02ee0dea..5a695e61c4e79a456c93df2e71600fc1a67e1f76 100644 (file)
--- a/core/mixer/audio/audio_mixer.cpp
+++ b/core/mixer/audio/audio_mixer.cpp
@@ -110,7 +110,7 @@ public:
                                prev = it->second;\r
                                \r
                        next_frame_transforms[item.tag] = next; // Store all active tags, inactive tags will be removed at the end.\r
-                               \r
+\r
                        if(next.volume < 0.001 && prev.volume < 0.001)\r
                                continue;\r
                                                                        \r
@@ -123,68 +123,54 @@ public:
                        const float prev_volume = static_cast<float>(prev.volume);\r
                        const float next_volume = static_cast<float>(next.volume);\r
                        const float delta               = 1.0f/static_cast<float>(format_desc_.audio_samples_per_frame/format_desc_.audio_channels);\r
-                       \r
-                       tbb::parallel_for\r
-                       (\r
-                               tbb::blocked_range<size_t>(0, format_desc_.audio_samples_per_frame/4),\r
-                               [&](const tbb::blocked_range<size_t>& r)\r
-                               {                                       \r
-                                       auto alpha_ps   = _mm_setr_ps(delta, delta, 0.0f, 0.0f);\r
-                                       auto delta2_ps  = _mm_set_ps1(delta*2.0f);\r
-                                       auto prev_ps    = _mm_set_ps1(prev_volume);\r
-                                       auto next_ps    = _mm_set_ps1(next_volume);     \r
-\r
-                                       for(size_t n = r.begin(); n < r.end(); ++n)\r
-                                       {               \r
-                                               auto next2_ps           = _mm_mul_ps(next_ps, alpha_ps);\r
-                                               auto prev2_ps           = _mm_sub_ps(prev_ps, _mm_mul_ps(prev_ps, alpha_ps));\r
-                                               auto volume_ps          = _mm_add_ps(next2_ps, prev2_ps);\r
-\r
-                                               auto sample_ps          = _mm_cvtepi32_ps(_mm_load_si128(reinterpret_cast<__m128i*>(&item.audio_data[n*4])));\r
-                                               auto res_sample_ps      = _mm_load_ps(&intermediate[n*4]);                                                                                      \r
-                                               sample_ps                       = _mm_mul_ps(sample_ps, volume_ps);     \r
-                                               res_sample_ps           = _mm_add_ps(sample_ps, res_sample_ps); \r
-\r
-                                               alpha_ps                        = _mm_add_ps(alpha_ps, delta2_ps);\r
-\r
-                                               _mm_store_ps(&intermediate[n*4], res_sample_ps);\r
-                                       }\r
-                               }\r
-                       , ap);\r
+                                               \r
+                       auto alpha_ps   = _mm_setr_ps(delta, delta, 0.0f, 0.0f);\r
+                       auto delta2_ps  = _mm_set_ps1(delta*2.0f);\r
+                       auto prev_ps    = _mm_set_ps1(prev_volume);\r
+                       auto next_ps    = _mm_set_ps1(next_volume);     \r
+\r
+                       for(size_t n = 0; n < format_desc_.audio_samples_per_frame/4; ++n)\r
+                       {               \r
+                               auto next2_ps           = _mm_mul_ps(next_ps, alpha_ps);\r
+                               auto prev2_ps           = _mm_sub_ps(prev_ps, _mm_mul_ps(prev_ps, alpha_ps));\r
+                               auto volume_ps          = _mm_add_ps(next2_ps, prev2_ps);\r
+\r
+                               auto sample_ps          = _mm_cvtepi32_ps(_mm_load_si128(reinterpret_cast<__m128i*>(&item.audio_data[n*4])));\r
+                               auto res_sample_ps      = _mm_load_ps(&intermediate[n*4]);                                                                                      \r
+                               sample_ps                       = _mm_mul_ps(sample_ps, volume_ps);     \r
+                               res_sample_ps           = _mm_add_ps(sample_ps, res_sample_ps); \r
+\r
+                               alpha_ps                        = _mm_add_ps(alpha_ps, delta2_ps);\r
+\r
+                               _mm_store_ps(&intermediate[n*4], res_sample_ps);\r
+                       }\r
                }\r
                \r
-               auto result = audio_buffer(format_desc_.audio_samples_per_frame+128, 0);        \r
-               \r
-               tbb::parallel_for\r
-               (\r
-                       tbb::blocked_range<size_t>(0, format_desc_.audio_samples_per_frame/32),\r
-                       [&](const tbb::blocked_range<size_t>& r)\r
-                       {               \r
-                               auto intermediate_128 = reinterpret_cast<__m128i*>(intermediate.data()+r.begin()*32);\r
-                               auto result_128           = reinterpret_cast<__m128i*>(result.data()+r.begin()*32);\r
+               auto result = audio_buffer(format_desc_.audio_samples_per_frame+128);   \r
+                       \r
+               auto intermediate_128 = reinterpret_cast<__m128i*>(intermediate.data());\r
+               auto result_128           = reinterpret_cast<__m128i*>(result.data());\r
                                \r
-                               for(size_t n = r.begin(); n < r.end(); ++n)\r
-                               {       \r
-                                       auto xmm0 = _mm_load_ps(reinterpret_cast<float*>(intermediate_128++));\r
-                                       auto xmm1 = _mm_load_ps(reinterpret_cast<float*>(intermediate_128++));\r
-                                       auto xmm2 = _mm_load_ps(reinterpret_cast<float*>(intermediate_128++));\r
-                                       auto xmm3 = _mm_load_ps(reinterpret_cast<float*>(intermediate_128++));\r
-                                       auto xmm4 = _mm_load_ps(reinterpret_cast<float*>(intermediate_128++));\r
-                                       auto xmm5 = _mm_load_ps(reinterpret_cast<float*>(intermediate_128++));\r
-                                       auto xmm6 = _mm_load_ps(reinterpret_cast<float*>(intermediate_128++));\r
-                                       auto xmm7 = _mm_load_ps(reinterpret_cast<float*>(intermediate_128++));\r
+               for(size_t n = 0; n < format_desc_.audio_samples_per_frame/32; ++n)\r
+               {       \r
+                       auto xmm0 = _mm_load_ps(reinterpret_cast<float*>(intermediate_128++));\r
+                       auto xmm1 = _mm_load_ps(reinterpret_cast<float*>(intermediate_128++));\r
+                       auto xmm2 = _mm_load_ps(reinterpret_cast<float*>(intermediate_128++));\r
+                       auto xmm3 = _mm_load_ps(reinterpret_cast<float*>(intermediate_128++));\r
+                       auto xmm4 = _mm_load_ps(reinterpret_cast<float*>(intermediate_128++));\r
+                       auto xmm5 = _mm_load_ps(reinterpret_cast<float*>(intermediate_128++));\r
+                       auto xmm6 = _mm_load_ps(reinterpret_cast<float*>(intermediate_128++));\r
+                       auto xmm7 = _mm_load_ps(reinterpret_cast<float*>(intermediate_128++));\r
                        \r
-                                       _mm_stream_si128(result_128++, _mm_cvtps_epi32(xmm0));\r
-                                       _mm_stream_si128(result_128++, _mm_cvtps_epi32(xmm1));\r
-                                       _mm_stream_si128(result_128++, _mm_cvtps_epi32(xmm2));\r
-                                       _mm_stream_si128(result_128++, _mm_cvtps_epi32(xmm3));\r
-                                       _mm_stream_si128(result_128++, _mm_cvtps_epi32(xmm4));\r
-                                       _mm_stream_si128(result_128++, _mm_cvtps_epi32(xmm5));\r
-                                       _mm_stream_si128(result_128++, _mm_cvtps_epi32(xmm6));\r
-                                       _mm_stream_si128(result_128++, _mm_cvtps_epi32(xmm7));\r
-                               }\r
-                       }\r
-               , ap);\r
+                       _mm_stream_si128(result_128++, _mm_cvtps_epi32(xmm0));\r
+                       _mm_stream_si128(result_128++, _mm_cvtps_epi32(xmm1));\r
+                       _mm_stream_si128(result_128++, _mm_cvtps_epi32(xmm2));\r
+                       _mm_stream_si128(result_128++, _mm_cvtps_epi32(xmm3));\r
+                       _mm_stream_si128(result_128++, _mm_cvtps_epi32(xmm4));\r
+                       _mm_stream_si128(result_128++, _mm_cvtps_epi32(xmm5));\r
+                       _mm_stream_si128(result_128++, _mm_cvtps_epi32(xmm6));\r
+                       _mm_stream_si128(result_128++, _mm_cvtps_epi32(xmm7));\r
+               }\r
 \r
                items.clear();\r
                prev_frame_transforms_ = std::move(next_frame_transforms);      \r