fix: check --all-features option in CI and fix identified issues (#323)

* ci: add new step for build and test workflows with `--all-features` option

* fix: remove nightly feature `stdsimd` (does not exist anymore?)

* fix: update pointer type conversion for avx512 usage

this needs testing, the `*mut i64` conversions seem very weird to me, but
the original `*mut u8` also were...

* fix: remove default `avx512` feature

my mistake, I added it for the lsp to diagnostic correctly

* fix: add missing error handling to metis graph init

* fix: add correct option to CI

also, don't run tests with all features since there's no avx512 on regular
GH runners

Author: imrn99

.github/workflows/ci.yml

@@ -24,7 +24,7 @@ jobs:
       - uses: Swatinem/rust-cache@v2
       - name: Install dependencies
         run: sudo apt-get -y install libclang-dev libmetis-dev libscotch-dev
-      - name: Run tests
+      - name: Run tests (default features)
         run: cargo test --all
 
   fmt:
@@ -78,5 +78,7 @@ jobs:
       - uses: Swatinem/rust-cache@v2
       - name: Install dependencies
         run: sudo apt-get -y install libclang-dev libmetis-dev libscotch-dev
-      - name: Build crates
+      - name: Build crates (default features)
         run: cargo build --all-targets
+      - name: Build crates (all features)
+        run: cargo build --all-targets --all-features

coupe/src/algorithms/recursive_bisection.rs

@@ -251,18 +251,18 @@ fn reorder_split_avx512<const D: usize, W>(
 
             for coord in 0..D {
                 _mm256_mask_compressstoreu_ps(
-                    items.points[coord].as_mut_ptr().add(lw) as *mut u8,
+                    items.points[coord].as_mut_ptr().add(lw),
                     mask,
                     val[coord],
                 );
             }
             _mm512_mask_compressstoreu_epi64(
-                items.weights.as_mut_ptr().add(lw) as *mut u8,
+                items.weights.as_mut_ptr().add(lw) as *mut i64, // TODO: test
                 mask,
                 val_weights,
             );
             _mm512_mask_compressstoreu_epi64(
-                items.parts.as_mut_ptr().add(lw) as *mut u8,
+                items.parts.as_mut_ptr().add(lw) as *mut i64, // TODO: test
                 mask,
                 val_parts,
             );
@@ -271,18 +271,18 @@ fn reorder_split_avx512<const D: usize, W>(
             rw -= nb_high;
             for coord in 0..D {
                 _mm256_mask_compressstoreu_ps(
-                    items.points[coord].as_mut_ptr().add(rw) as *mut u8,
+                    items.points[coord].as_mut_ptr().add(rw),
                     !mask,
                     val[coord],
                 );
             }
             _mm512_mask_compressstoreu_epi64(
-                items.weights.as_mut_ptr().add(rw) as *mut u8,
+                items.weights.as_mut_ptr().add(rw) as *mut i64, // TODO: test
                 !mask,
                 val_weights,
             );
             _mm512_mask_compressstoreu_epi64(
-                items.parts.as_mut_ptr().add(rw) as *mut u8,
+                items.parts.as_mut_ptr().add(rw) as *mut i64, // TODO: test
                 !mask,
                 val_parts,
             );
@@ -300,37 +300,37 @@ fn reorder_split_avx512<const D: usize, W>(
         let nb_high = mask_high.count_ones() as usize;
         for coord in 0..D {
             _mm256_mask_compressstoreu_ps(
-                items.points[coord].as_mut_ptr().add(lw) as *mut u8,
+                items.points[coord].as_mut_ptr().add(lw),
                 mask_low,
                 val[coord],
             );
         }
         _mm512_mask_compressstoreu_epi64(
-            items.weights.as_mut_ptr().add(lw) as *mut u8,
+            items.weights.as_mut_ptr().add(lw) as *mut i64, // TODO: test
             mask_low,
             val_weights,
         );
         _mm512_mask_compressstoreu_epi64(
-            items.parts.as_mut_ptr().add(lw) as *mut u8,
+            items.parts.as_mut_ptr().add(lw) as *mut i64, // TODO: test
             mask_low,
             val_parts,
         );
         lw += nb_low;
         rw -= nb_high;
         for coord in 0..D {
             _mm256_mask_compressstoreu_ps(
-                items.points[coord].as_mut_ptr().add(rw) as *mut u8,
+                items.points[coord].as_mut_ptr().add(rw),
                 mask_high,
                 val[coord],
             );
         }
         _mm512_mask_compressstoreu_epi64(
-            items.weights.as_mut_ptr().add(rw) as *mut u8,
+            items.weights.as_mut_ptr().add(rw) as *mut i64, // TODO: test
             mask_high,
             val_weights,
         );
         _mm512_mask_compressstoreu_epi64(
-            items.parts.as_mut_ptr().add(rw) as *mut u8,
+            items.parts.as_mut_ptr().add(rw) as *mut i64, // TODO: test
             mask_high,
             val_parts,
         );
@@ -340,37 +340,37 @@ fn reorder_split_avx512<const D: usize, W>(
         let nb_high = AVX2_REGISTER_BYTES - nb_low;
         for coord in 0..D {
             _mm256_mask_compressstoreu_ps(
-                items.points[coord].as_mut_ptr().add(lw) as *mut u8,
+                items.points[coord].as_mut_ptr().add(lw),
                 mask,
                 lv[coord],
             );
         }
         _mm512_mask_compressstoreu_epi64(
-            items.weights.as_mut_ptr().add(lw) as *mut u8,
+            items.weights.as_mut_ptr().add(lw) as *mut i64, // TODO: test
             mask,
             lv_weights,
         );
         _mm512_mask_compressstoreu_epi64(
-            items.parts.as_mut_ptr().add(lw) as *mut u8,
+            items.parts.as_mut_ptr().add(lw) as *mut i64, // TODO: test
             mask,
             lv_parts,
         );
         lw += nb_low;
         rw -= nb_high;
         for coord in 0..D {
             _mm256_mask_compressstoreu_ps(
-                items.points[coord].as_mut_ptr().add(rw) as *mut u8,
+                items.points[coord].as_mut_ptr().add(rw),
                 !mask,
                 lv[coord],
             );
         }
         _mm512_mask_compressstoreu_epi64(
-            items.weights.as_mut_ptr().add(rw) as *mut u8,
+            items.weights.as_mut_ptr().add(rw) as *mut i64, // TODO: test
             !mask,
             lv_weights,
         );
         _mm512_mask_compressstoreu_epi64(
-            items.parts.as_mut_ptr().add(rw) as *mut u8,
+            items.parts.as_mut_ptr().add(rw) as *mut i64, // TODO: test
             !mask,
             lv_parts,
         );
@@ -380,37 +380,37 @@ fn reorder_split_avx512<const D: usize, W>(
         let nb_high = AVX2_REGISTER_BYTES - nb_low;
         for coord in 0..D {
             _mm256_mask_compressstoreu_ps(
-                items.points[coord].as_mut_ptr().add(lw) as *mut u8,
+                items.points[coord].as_mut_ptr().add(lw),
                 mask,
                 rv[coord],
             );
         }
         _mm512_mask_compressstoreu_epi64(
-            items.weights.as_mut_ptr().add(lw) as *mut u8,
+            items.weights.as_mut_ptr().add(lw) as *mut i64, // TODO: test
             mask,
             rv_weights,
         );
         _mm512_mask_compressstoreu_epi64(
-            items.parts.as_mut_ptr().add(lw) as *mut u8,
+            items.parts.as_mut_ptr().add(lw) as *mut i64, // TODO: test
             mask,
             rv_parts,
         );
         lw += nb_low;
         rw -= nb_high;
         for coord in 0..D {
             _mm256_mask_compressstoreu_ps(
-                items.points[coord].as_mut_ptr().add(rw) as *mut u8,
+                items.points[coord].as_mut_ptr().add(rw),
                 !mask,
                 rv[coord],
             );
         }
         _mm512_mask_compressstoreu_epi64(
-            items.weights.as_mut_ptr().add(rw) as *mut u8,
+            items.weights.as_mut_ptr().add(rw) as *mut i64, // TODO: test
             !mask,
             rv_weights,
         );
         _mm512_mask_compressstoreu_epi64(
-            items.parts.as_mut_ptr().add(rw) as *mut u8,
+            items.parts.as_mut_ptr().add(rw) as *mut i64, // TODO: test
             !mask,
             rv_parts,
         );

coupe/src/lib.rs

@@ -29,7 +29,6 @@
 //! - [Fiduccia-Mattheyses][FiducciaMattheyses]
 //! - [Kernighan-Lin][KernighanLin]
 
-#![cfg_attr(feature = "avx512", feature(stdsimd))]
 #![warn(
     missing_copy_implementations,
     missing_debug_implementations,

tools/lib/metis.rs

@@ -35,7 +35,7 @@ impl<const D: usize> ToRunner<D> for Recursive {
 
         let mut metis_partition = vec![0; problem.adjacency().rows()];
         Box::new(move |partition| {
-            let mut graph = metis::Graph::new(ncon, self.part_count, &mut xadj, &mut adjncy);
+            let mut graph = metis::Graph::new(ncon, self.part_count, &mut xadj, &mut adjncy)?;
             if let Some(weights) = &mut weights {
                 graph = graph.set_vwgt(weights);
             }
@@ -85,7 +85,7 @@ impl<const D: usize> ToRunner<D> for KWay {
 
         let mut metis_partition = vec![0; problem.adjacency().rows()];
         Box::new(move |partition| {
-            let mut graph = metis::Graph::new(ncon, self.part_count, &mut xadj, &mut adjncy);
+            let mut graph = metis::Graph::new(ncon, self.part_count, &mut xadj, &mut adjncy)?;
             if let Some(weights) = &mut weights {
                 graph = graph.set_vwgt(weights);
             }