Merge pull request #3024 from autonomys/gpu-plotting-improvements

GPU plotting improvements

crates/subspace-farmer/src/bin/subspace-farmer/commands/cluster/plotter.rs

@@ -47,6 +47,8 @@ struct CpuPlottingOptions {
     /// `--cpu-sector-downloading-concurrency` and setting this option higher than
     /// `--cpu-sector-downloading-concurrency` will have no effect.
     ///
+    /// CPU plotting is disabled by default if GPU plotting is detected.
+    ///
     /// Increase will result in higher memory usage, setting to 0 will disable CPU plotting.
     #[arg(long)]
     cpu_sector_encoding_concurrency: Option<usize>,
@@ -151,34 +153,42 @@ where
     let mut legacy_plotters = Vec::<Box<dyn Plotter + Send + Sync>>::new();
     let mut modern_plotters = Vec::<Box<dyn Plotter + Send + Sync>>::new();
 
+    #[cfg(feature = "cuda")]
     {
-        let maybe_cpu_plotters = init_cpu_plotters::<_, PosTableLegacy, PosTable>(
-            cpu_plotting_options,
+        let maybe_cuda_plotter = init_cuda_plotter(
+            cuda_plotting_options,
             piece_getter.clone(),
             Arc::clone(&global_mutex),
             kzg.clone(),
             erasure_coding.clone(),
             registry,
         )?;
 
-        if let Some((legacy_cpu_plotter, modern_cpu_plotter)) = maybe_cpu_plotters {
-            legacy_plotters.push(Box::new(legacy_cpu_plotter));
-            modern_plotters.push(Box::new(modern_cpu_plotter));
+        if let Some(cuda_plotter) = maybe_cuda_plotter {
+            modern_plotters.push(Box::new(cuda_plotter));
         }
     }
-    #[cfg(feature = "cuda")]
     {
-        let maybe_cuda_plotter = init_cuda_plotter(
-            cuda_plotting_options,
+        let cpu_sector_encoding_concurrency = cpu_plotting_options.cpu_sector_encoding_concurrency;
+        let maybe_cpu_plotters = init_cpu_plotters::<_, PosTableLegacy, PosTable>(
+            cpu_plotting_options,
             piece_getter,
             global_mutex,
             kzg,
             erasure_coding,
             registry,
         )?;
 
-        if let Some(cuda_plotter) = maybe_cuda_plotter {
-            modern_plotters.push(Box::new(cuda_plotter));
+        if let Some((legacy_cpu_plotter, modern_cpu_plotter)) = maybe_cpu_plotters {
+            legacy_plotters.push(Box::new(legacy_cpu_plotter));
+            if !modern_plotters.is_empty() && cpu_sector_encoding_concurrency.is_none() {
+                info!(
+                    "CPU plotting for v1 farms was disabled due to detected faster plotting with \
+                    GPU"
+                );
+            } else {
+                modern_plotters.push(Box::new(modern_cpu_plotter));
+            }
         }
     }
     let legacy_plotter = Arc::new(PoolPlotter::new(legacy_plotters, PLOTTING_RETRY_INTERVAL));

crates/subspace-farmer/src/bin/subspace-farmer/commands/farm.rs

@@ -83,6 +83,8 @@ struct CpuPlottingOptions {
     /// `--cpu-sector-downloading-concurrency` and setting this option higher than
     /// `--cpu-sector-downloading-concurrency` will have no effect.
     ///
+    /// CPU plotting is disabled by default if GPU plotting is detected.
+    ///
     /// Increase will result in higher memory usage, setting to 0 will disable CPU plotting.
     #[arg(long)]
     cpu_sector_encoding_concurrency: Option<usize>,
@@ -469,34 +471,42 @@ where
     let mut legacy_plotters = Vec::<Box<dyn Plotter + Send + Sync>>::new();
     let mut modern_plotters = Vec::<Box<dyn Plotter + Send + Sync>>::new();
 
+    #[cfg(feature = "cuda")]
     {
-        let maybe_cpu_plotters = init_cpu_plotters::<_, PosTableLegacy, PosTable>(
-            cpu_plotting_options,
+        let maybe_cuda_plotter = init_cuda_plotter(
+            cuda_plotting_options,
             piece_getter.clone(),
             Arc::clone(&global_mutex),
             kzg.clone(),
             erasure_coding.clone(),
             &mut registry,
         )?;
 
-        if let Some((legacy_cpu_plotter, modern_cpu_plotter)) = maybe_cpu_plotters {
-            legacy_plotters.push(Box::new(legacy_cpu_plotter));
-            modern_plotters.push(Box::new(modern_cpu_plotter));
+        if let Some(cuda_plotter) = maybe_cuda_plotter {
+            modern_plotters.push(Box::new(cuda_plotter));
         }
     }
-    #[cfg(feature = "cuda")]
     {
-        let maybe_cuda_plotter = init_cuda_plotter(
-            cuda_plotting_options,
+        let cpu_sector_encoding_concurrency = cpu_plotting_options.cpu_sector_encoding_concurrency;
+        let maybe_cpu_plotters = init_cpu_plotters::<_, PosTableLegacy, PosTable>(
+            cpu_plotting_options,
             piece_getter.clone(),
             Arc::clone(&global_mutex),
             kzg.clone(),
             erasure_coding.clone(),
             &mut registry,
         )?;
 
-        if let Some(cuda_plotter) = maybe_cuda_plotter {
-            modern_plotters.push(Box::new(cuda_plotter));
+        if let Some((legacy_cpu_plotter, modern_cpu_plotter)) = maybe_cpu_plotters {
+            legacy_plotters.push(Box::new(legacy_cpu_plotter));
+            if !modern_plotters.is_empty() && cpu_sector_encoding_concurrency.is_none() {
+                info!(
+                    "CPU plotting for v1 farms was disabled due to detected faster plotting with \
+                    GPU"
+                );
+            } else {
+                modern_plotters.push(Box::new(modern_cpu_plotter));
+            }
         }
     }
     let legacy_plotter = Arc::new(PoolPlotter::new(legacy_plotters, PLOTTING_RETRY_INTERVAL));

shared/subspace-proof-of-space-gpu/src/cuda.rs

@@ -112,7 +112,7 @@ impl CudaDevice {
         let mut challenge_index_gpu = Vec::<u32>::with_capacity(challenge_len);
         let mut parity_record_chunks = Vec::<Scalar>::with_capacity(Record::NUM_CHUNKS);
 
-        let err = unsafe {
+        let error = unsafe {
             generate_and_encode_pospace_dispatch(
                 u32::from(PosProof::K),
                 &**seed,
@@ -126,8 +126,15 @@ impl CudaDevice {
             )
         };
 
-        if err.code != 0 {
-            return Err(String::from(err));
+        if error.code != 0 {
+            let error = error.to_string();
+            if error.contains("the provided PTX was compiled with an unsupported toolchain.") {
+                return Err(format!(
+                    "Nvidia driver is likely too old, make sure install version 550 or newer: \
+                    {error}"
+                ));
+            }
+            return Err(error);
         }
 
         let proof_count = proof_count as usize;