Changeset c97f7ed

Timestamp:

Dec 29, 2021, 5:51:48 PM (2 years ago)

Author:

Paul Brossier <piem@piem.org>

Branches:

feature/cnn, feature/crepe

Children:

d91fa83

Parents:

337e70d

git-author:

Paul Brossier <piem@piem.org> (01/08/19 15:45:59)

git-committer:

Paul Brossier <piem@piem.org> (12/29/21 17:51:48)

Message:

[conv1d] add second blas optimisation using gemv

File:

• src/ai/conv1d.c (modified) (2 diffs)

src/ai/conv1d.c

@@ -289 +289 @@
 
   uint_t sdot_size = c->kernel->shape[0] * c->kernel->shape[1];
-  uint_t input_stride = c->stride_shape * c->kernel->shape[1];
+  uint_t input_stride = c->stride_shape * c->padded_input->shape[1]
 
   AUBIO_ASSERT(c && input_tensor && activations);
@@ -308 +308 @@
   // for each output
   for (j = 0; j < activations->shape[0]; j++) {
+    // for each row of activation output
+    aubio_cblas__gemv(CblasRowMajor, CblasTrans,
+        sdot_size, c->kernel->shape[2], 1.,
+        c->kernel->buffer, c->kernel->shape[2],
+        c->padded_input->buffer + j  * input_stride, 1, 0.,
+        activations->buffer + j * activations->shape[1], 1);
+  }
+  for (j = 0; j < activations->shape[0]; j++) {
     // for each kernel filter k
     for (i = 0; i < activations->shape[1]; i++) {
-      // get bias
-      bias = c->bias->data[i];
-
-      // compute one activation output
-      acc = aubio_cblas_dot(sdot_size, c->kernel->buffer + i,
-          c->kernel->shape[2], c->padded_input->buffer + j * input_stride, 1);
-
-      // apply bias
-      acc += bias;
-
-      // compute RELU
-      activations->data[j][i] = MAX(acc, 0.);
+      activations->data[j][i] += c->bias->data[i];
+      activations->data[j][i] = MAX(activations->data[j][i], 0);
     }
   }