More optimizations

• A lot of misc. optimization work, notably parallelization of Rodinia CPU schedules.
• Support >32 tiles in GPU backend.
• Tweak some transformations.

juno_samples/rodinia/cfd/src/lib.rs

@@ -19,6 +19,8 @@ pub struct CFDInputs {
     pub block_size: usize,
     #[clap(short = None, long = Some("pre-euler"))]
     pub pre_euler: bool,
+    #[clap(short, long)]
+    pub verify: bool,
 }
 
 fn run_euler(
@@ -219,6 +221,7 @@ pub fn cfd_harness(args: CFDInputs) {
         iterations,
         block_size,
         pre_euler,
+        verify,
     } = args;
 
     let FarFieldConditions {
@@ -268,37 +271,39 @@ pub fn cfd_harness(args: CFDInputs) {
             &ff_fc_momentum_z,
         )
     };
-    let res_rust = if pre_euler {
-        rust_cfd::pre_euler(
-            nelr,
-            iterations,
-            variables,
-            areas.as_slice(),
-            elements_surrounding_elements.as_slice(),
-            &normals,
-            &ff_variable,
-            &ff_fc_density_energy,
-            &ff_fc_momentum_x,
-            &ff_fc_momentum_y,
-            &ff_fc_momentum_z,
-        )
-    } else {
-        rust_cfd::euler(
-            nelr,
-            iterations,
-            variables,
-            areas.as_slice(),
-            elements_surrounding_elements.as_slice(),
-            &normals,
-            &ff_variable,
-            &ff_fc_density_energy,
-            &ff_fc_momentum_x,
-            &ff_fc_momentum_y,
-            &ff_fc_momentum_z,
-        )
-    };
+    if verify {
+        let res_rust = if pre_euler {
+            rust_cfd::pre_euler(
+                nelr,
+                iterations,
+                variables,
+                areas.as_slice(),
+                elements_surrounding_elements.as_slice(),
+                &normals,
+                &ff_variable,
+                &ff_fc_density_energy,
+                &ff_fc_momentum_x,
+                &ff_fc_momentum_y,
+                &ff_fc_momentum_z,
+            )
+        } else {
+            rust_cfd::euler(
+                nelr,
+                iterations,
+                variables,
+                areas.as_slice(),
+                elements_surrounding_elements.as_slice(),
+                &normals,
+                &ff_variable,
+                &ff_fc_density_energy,
+                &ff_fc_momentum_x,
+                &ff_fc_momentum_y,
+                &ff_fc_momentum_z,
+            )
+        };
 
-    if !compare_floats(&res_juno, &res_rust) {
-        panic!("Mismatch in results");
+        if !compare_floats(&res_juno, &res_rust) {
+            panic!("Mismatch in results");
+        }
     }
 }