Fix CI (#123)

tracel-ai · Sep 16, 2024 · 443b5c8 · 443b5c8
1 parent 73dada8
commit 443b5c8
Show file tree

Hide file tree

Showing 15 changed files with 12 additions and 7 deletions.
diff --git a/crates/cubecl-macros/tests/cuda/common.rs → crates/cubecl-cuda/tests/common.rs b/crates/cubecl-macros/tests/cuda/common.rs → crates/cubecl-cuda/tests/common.rs
diff --git a/crates/cubecl-macros/tests/cuda/main.rs → crates/cubecl-cuda/tests/main.rs b/crates/cubecl-macros/tests/cuda/main.rs → crates/cubecl-cuda/tests/main.rs
diff --git a/...cl-macros/tests/cuda/sequence_for_loop.cu → ...es/cubecl-cuda/tests/sequence_for_loop.cu b/...cl-macros/tests/cuda/sequence_for_loop.cu → ...es/cubecl-cuda/tests/sequence_for_loop.cu
diff --git a/.../cubecl-macros/tests/cuda/slice_assign.cu → crates/cubecl-cuda/tests/slice_assign.cu b/.../cubecl-macros/tests/cuda/slice_assign.cu → crates/cubecl-cuda/tests/slice_assign.cu
diff --git a/...s/cubecl-macros/tests/cuda/subcube_sum.cu → crates/cubecl-cuda/tests/subcube_sum.cu b/...s/cubecl-macros/tests/cuda/subcube_sum.cu → crates/cubecl-cuda/tests/subcube_sum.cu
diff --git a/...s/cubecl-macros/tests/cuda/unary_bench.cu → crates/cubecl-cuda/tests/unary_bench.cu b/...s/cubecl-macros/tests/cuda/unary_bench.cu → crates/cubecl-cuda/tests/unary_bench.cu
diff --git a/crates/cubecl-linalg/src/matmul/tests/cmma/load_shared_memory.rs b/crates/cubecl-linalg/src/matmul/tests/cmma/load_shared_memory.rs
@@ -32,6 +32,8 @@ fn load_lhs_test<F: Float>(
         lhs_sm[i] = F::new(0.);
     }
 
+    sync_units();
+
     let offsets = Offsets {
         batch_lhs: 0,
         batch_rhs: 0,
@@ -48,6 +50,8 @@ fn load_lhs_test<F: Float>(
 
     load_lhs(lhs_tensor, &mut lhs_sm, 2, k_offset, runtime_info, config);
 
+    sync_units();
+
     for i in 0..sm_size {
         lhs_sm_arr[i] = lhs_sm[i];
     }
@@ -72,6 +76,8 @@ fn load_rhs_test<F: Float>(
         rhs_sm[i] = F::new(0.);
     }
 
+    sync_units();
+
     let offsets = Offsets {
         batch_lhs: 0,
         batch_rhs: 0,
@@ -88,6 +94,8 @@ fn load_rhs_test<F: Float>(
 
     load_rhs(rhs_tensor, &mut rhs_sm, 2, k_offset, runtime_info, config);
 
+    sync_units();
+
     for i in 0..sm_size {
         rhs_sm_arr[i] = rhs_sm[i];
     }

diff --git a/crates/cubecl-linalg/src/matmul/tests/tiling2d/load_shared_memory.rs b/crates/cubecl-linalg/src/matmul/tests/tiling2d/load_shared_memory.rs
@@ -170,6 +170,8 @@ fn load_tensor_multiple_tiles_test<F: Float>(
         shared_memory[i] = F::vectorized(0., tile_size);
     }
 
+    sync_units();
+
     let unit_row = 4 * UNIT_POS_X;
     let unit_col = 4 * UNIT_POS_Y;
     let batch_offset = 0;
@@ -213,6 +215,8 @@ fn load_tensor_multiple_tiles_test<F: Float>(
         load_rhs_plain::<F, TileLoader<F>>(tensor, info, config);
     }
 
+    sync_units();
+
     for i in 0..sm_size {
         sm_out[i] = shared_memory[i];
     }

diff --git a/crates/cubecl-macros/Cargo.toml b/crates/cubecl-macros/Cargo.toml
@@ -30,10 +30,3 @@ quote = { workspace = true }
 syn = { workspace = true }
 
 cubecl-common = { path = "../cubecl-common", version = "0.2", default-features = false }
-
-[dev-dependencies]
-cubecl-core = { path = "../cubecl-core", version = "0.2", default-features = false }
-cubecl-cuda = { path = "../cubecl-cuda", version = "0.2", default-features = true }
-cubecl-linalg = { path = "../cubecl-linalg", version = "0.2", default-features = false }
-cubecl-wgpu = { path = "../cubecl-wgpu", version = "0.2", default-features = false }
-pretty_assertions = { workspace = true }
diff --git a/crates/cubecl-macros/tests/wgpu/common.rs → crates/cubecl-wgpu/tests/common.rs b/crates/cubecl-macros/tests/wgpu/common.rs → crates/cubecl-wgpu/tests/common.rs
diff --git a/crates/cubecl-macros/tests/wgpu/main.rs → crates/cubecl-wgpu/tests/main.rs b/crates/cubecl-macros/tests/wgpu/main.rs → crates/cubecl-wgpu/tests/main.rs
diff --git a/...-macros/tests/wgpu/sequence_for_loop.wgsl → .../cubecl-wgpu/tests/sequence_for_loop.wgsl b/...-macros/tests/wgpu/sequence_for_loop.wgsl → .../cubecl-wgpu/tests/sequence_for_loop.wgsl
diff --git a/...ubecl-macros/tests/wgpu/slice_assign.wgsl → crates/cubecl-wgpu/tests/slice_assign.wgsl b/...ubecl-macros/tests/wgpu/slice_assign.wgsl → crates/cubecl-wgpu/tests/slice_assign.wgsl
diff --git a/...cubecl-macros/tests/wgpu/subcube_sum.wgsl → crates/cubecl-wgpu/tests/subcube_sum.wgsl b/...cubecl-macros/tests/wgpu/subcube_sum.wgsl → crates/cubecl-wgpu/tests/subcube_sum.wgsl
diff --git a/...cubecl-macros/tests/wgpu/unary_bench.wgsl → crates/cubecl-wgpu/tests/unary_bench.wgsl b/...cubecl-macros/tests/wgpu/unary_bench.wgsl → crates/cubecl-wgpu/tests/unary_bench.wgsl