Merge pull request #834 from anarkiwi/uhds

VkFFT performance improvements for both worker and scanner.
IQTLabs · Aug 22, 2023 · fe44725 · fe44725
2 parents 099b54d + 8ac94dd
commit fe44725
Show file tree

Hide file tree

Showing 2 changed files with 16 additions and 8 deletions.
diff --git a/docker/Dockerfile.base b/docker/Dockerfile.base
@@ -38,7 +38,7 @@ RUN cmake .. && make -j "$(nproc)"
 FROM iqtlabs/gnuradio:3.10.7 as gr-iqtlabs-builder
 ENV DEBIAN_FRONTEND noninteractive
 WORKDIR /root
-RUN git clone https://github.com/iqtlabs/gr-iqtlabs -b 1.0.33
+RUN git clone https://github.com/iqtlabs/gr-iqtlabs -b 1.0.34
 RUN apt-get update && apt-get install -y --no-install-recommends libopencv-dev libvulkan-dev
 WORKDIR /root/gr-iqtlabs/build
 COPY --from=sigmf-builder /usr/local /usr/local
@@ -48,7 +48,7 @@ RUN cmake .. && make -j "$(nproc)" && make install
 FROM iqtlabs/gnuradio:3.10.7 as uhd_sample_recorder-builder
 ENV DEBIAN_FRONTEND noninteractive
 WORKDIR /root
-RUN git clone https://github.com/iqtlabs/uhd_sample_recorder -b v1.0.5
+RUN git clone https://github.com/iqtlabs/uhd_sample_recorder -b v1.0.6
 WORKDIR /root/uhd_sample_recorder
 RUN ./bin/install-deps.sh && ./bin/build.sh
 WORKDIR /root/uhd_sample_recorder/build

diff --git a/gamutrf/grscan.py b/gamutrf/grscan.py
@@ -260,24 +260,26 @@ def get_db_blocks(self, fft_size, samp_rate, scaling):
             raise ValueError("scaling must be 'spectrum' or 'density'")
         return [
             blocks.complex_to_mag_squared(fft_size),
-            blocks.multiply_const_vff([scale] * fft_size),
+            blocks.multiply_const_ff(scale, fft_size),
             blocks.nlog10_ff(10, fft_size, 0),
         ]
 
     def get_window(self, fft_size):
         return window.hann(fft_size)
 
-    def get_offload_fft_block(self, fft_batch_size, fft_size, fft_block):
-        return [
+    def get_offload_fft_block(self, fft_batch_size, fft_size, fft_block, fft_roll):
+        offload_blocks = [
             blocks.stream_to_vector(gr.sizeof_gr_complex, fft_batch_size * fft_size),
             blocks.multiply_const_vff(
                 [val for val in self.get_window(fft_size) for _ in range(2)]
                 * fft_batch_size
             ),
             fft_block,
             blocks.vector_to_stream(gr.sizeof_gr_complex * fft_size, fft_batch_size),
-            self.iqtlabs.vector_roll(fft_size),
         ]
+        if fft_roll:
+            offload_blocks.append(self.iqtlabs.vector_roll(fft_size)),
+        return offload_blocks
 
     def get_fft_blocks(
         self, use_vkfft, fft_batch_size, fft_size, dc_block_len, dc_block_long
@@ -286,15 +288,21 @@ def get_fft_blocks(
         if dc_block_len:
             fft_blocks.append(grfilter.dc_blocker_cc(dc_block_len, dc_block_long))
         fft_block = None
+        fft_roll = False
         if self.wavelearner:
             fft_block = self.wavelearner.fft(
                 int(fft_batch_size * fft_size), (fft_size), True
             )
+            fft_roll = True
         elif use_vkfft:
-            fft_block = self.iqtlabs.vkfft(int(fft_batch_size * fft_size), fft_size)
+            fft_block = self.iqtlabs.vkfft(
+                int(fft_batch_size * fft_size), fft_size, True
+            )
         if fft_block:
             fft_blocks.extend(
-                self.get_offload_fft_block(fft_batch_size, fft_size, fft_block)
+                self.get_offload_fft_block(
+                    fft_batch_size, fft_size, fft_block, fft_roll
+                )
             )
         else:
             fft_blocks.extend(