removed redundant nvtx ranges

Netflix · Mar 12, 2024 · a8286da · a8286da
1 parent c6a7a92
commit a8286da
Show file tree

Hide file tree

Showing 3 changed files with 27 additions and 14 deletions.
diff --git a/libvmaf/src/feature/cuda/integer_adm_cuda.c b/libvmaf/src/feature/cuda/integer_adm_cuda.c
@@ -32,7 +32,9 @@
 #include "picture_cuda.h"
 #include <unistd.h>
 #include <assert.h>
+#ifdef HAVE_NVTX
 #include "nvtx3/nvToolsExt.h"
+#endif
 
 #define RES_BUFFER_SIZE 4 * 3 * 2
 
@@ -642,7 +644,6 @@ typedef struct write_score_parameters_adm {
 
 static int write_scores(write_score_parameters_adm* params)
 {
-    nvtxRangePushA("write_scores ADM");
     VmafFeatureCollector *feature_collector = params->feature_collector;
     AdmStateCuda *s = params->s;
     unsigned index = params->index;
@@ -717,7 +718,6 @@ static int write_scores(write_score_parameters_adm* params)
 
     if (!s->debug) {
 
-        nvtxRangePop();
         return err;
     }
 
@@ -754,7 +754,6 @@ static int write_scores(write_score_parameters_adm* params)
 
     err |= vmaf_feature_collector_append_with_dict(feature_collector,
             s->feature_name_dict, "integer_adm_den_scale3", scores[7], index);
-    nvtxRangePop();
     return err;
 }
 
@@ -1233,11 +1232,17 @@ static int close_fex_cuda(VmafFeatureExtractor *fex)
 static int flush_fex_cuda(VmafFeatureExtractor *fex,
         VmafFeatureCollector *feature_collector)
 {
+#ifdef HAVE_NVTX
+    nvtxRangePushA("flush adm_cuda");
+#endif
     AdmStateCuda *s = fex->priv;
     int ret = 0;
     CHECK_CUDA(cuStreamSynchronize(s->str));
     CHECK_CUDA(cuStreamSynchronize(s->host_stream));
     CHECK_CUDA(cuEventSynchronize(s->scores_written));
+#ifdef HAVE_NVTX
+    nvtxRangePop();
+#endif
     return (ret < 0) ? ret : !ret;
 }
 

diff --git a/libvmaf/src/feature/cuda/integer_motion_cuda.c b/libvmaf/src/feature/cuda/integer_motion_cuda.c
@@ -31,7 +31,9 @@
 #include "picture.h"
 #include "picture_cuda.h"
 #include "cuda_helper.cuh"
+#ifdef HAVE_NVTX
 #include "nvtx3/nvToolsExt.h"
+#endif
 
 typedef struct MotionStateCuda {
     CUevent event, finished, scores_written;
@@ -212,7 +214,9 @@ static int init_fex_cuda(VmafFeatureExtractor *fex, enum VmafPixelFormat pix_fmt
 static int flush_fex_cuda(VmafFeatureExtractor *fex,
         VmafFeatureCollector *feature_collector)
 {
-    nvtxRangePushA("FLUSH MOT");
+#ifdef HAVE_NVTX
+    nvtxRangePushA("flush motion_cuda");
+#endif
 
     MotionStateCuda *s = fex->priv;
     int ret = 0;
@@ -224,7 +228,7 @@ static int flush_fex_cuda(VmafFeatureExtractor *fex,
             continue;
         }
         CHECK_CUDA(cuEventSynchronize(s->scores_written));
-        nvtxRangePop();
+
     } 
     else {
         if (s->index > 0 && !s->closed) {
@@ -235,6 +239,10 @@ static int flush_fex_cuda(VmafFeatureExtractor *fex,
         s->closed = true;
     }
     s->flushed = true;
+
+#ifdef HAVE_NVTX
+        nvtxRangePop();
+#endif
     return (ret < 0) ? ret : !ret;
 }
 

diff --git a/libvmaf/src/feature/cuda/integer_vif_cuda.c b/libvmaf/src/feature/cuda/integer_vif_cuda.c
@@ -33,8 +33,9 @@
 #include "cuda/integer_vif_cuda.h"
 #include "picture_cuda.h"
 
-
+#ifdef HAVE_NVTX
 #include "nvtx3/nvToolsExt.h"
+#endif
 
 #if ARCH_X86
 #include "x86/vif_avx2.h"
@@ -349,7 +350,6 @@ typedef struct VifScore {
 
 static int write_scores(write_score_parameters_vif* data)
 {
-    nvtxRangePushA("write_scoes VIF");
     VmafFeatureCollector *feature_collector = data->feature_collector;
     VifStateCuda *s = data->s;
     unsigned index = data->index;
@@ -384,11 +384,7 @@ static int write_scores(write_score_parameters_vif* data)
             s->feature_name_dict, "VMAF_integer_feature_vif_scale3_score",
             vif.scale[3].num / vif.scale[3].den, index);
 
-    if (!s->debug) {
-
-        nvtxRangePop();
-        return err;
-    }
+    if (!s->debug) return err;
 
     const double score_num =
         (double)vif.scale[0].num + (double)vif.scale[1].num +
@@ -441,7 +437,7 @@ static int write_scores(write_score_parameters_vif* data)
     err |= vmaf_feature_collector_append_with_dict(feature_collector,
             s->feature_name_dict, "integer_vif_den_scale3", vif.scale[3].den,
             index);
-    nvtxRangePop();
+
     return err;
 }
 
@@ -536,14 +532,18 @@ static int close_fex_cuda(VmafFeatureExtractor *fex)
 static int flush_fex_cuda(VmafFeatureExtractor *fex,
         VmafFeatureCollector *feature_collector)
 {
-    nvtxRangePushA("flush VIF");
+#ifdef HAVE_NVTX
+    nvtxRangePushA("flush vif_cuda");
+#endif
     VifStateCuda *s = fex->priv;
 
     int ret = 0;
     CHECK_CUDA(cuStreamSynchronize(s->str));
     CHECK_CUDA(cuStreamSynchronize(s->host_stream));
     CHECK_CUDA(cuEventSynchronize(s->scores_written));
+#ifdef HAVE_NVTX
     nvtxRangePop();
+#endif
 
     return (ret < 0) ? ret : !ret;
 }