Merge branch 'main' into parquet-no-default-features

Author: ivarflakstad

Cargo.toml

@@ -59,7 +59,7 @@ cudarc = { version = "0.19.8", features = [
     "cuda-version-from-build-system",
     "dynamic-linking",
 ], default-features = false }
-cutile = "=0.3.0"
+cutile = "=0.3.1"
 fancy-regex = "0.19.0"
 gemm = { version = "0.19.0", features = ["wasm-simd128-enable"] }
 hf-hub = { version = "1.0.0", features = ["blocking"] }

candle-book/src/guide/cutile.md

@@ -44,7 +44,6 @@ use cutile::tile_kernel::TileKernel;
 mod kernels {
     use candle_core::cutile;
     use cutile::core::*;
-    use cutile::cutile_compiler;
 
     // Define #[cutile::entry] kernels here.
 }

candle-core/examples/cutile.rs

@@ -9,7 +9,6 @@ const BLOCK_SIZE: usize = 32;
 mod kernels {
     use candle_core::cutile;
     use cutile::core::*;
-    use cutile::cutile_compiler;
 
     unsafe fn tensor(ptr: *mut f32, len: i32) -> Tensor<f32, { [-1] }> {
         let shape: Shape<{ [-1] }> = Shape::<{ [-1] }> { dims: &[len] };

candle-core/tests/cutile_tests.rs

@@ -16,7 +16,6 @@ static GPU_TEST_LOCK: Mutex<()> = Mutex::new(());
 mod kernels {
     use candle_core::cutile;
     use cutile::core::*;
-    use cutile::cutile_compiler;
 
     unsafe fn tensor(ptr: *mut f32, len: i32) -> Tensor<f32, { [-1] }> {
         let shape: Shape<{ [-1] }> = Shape::<{ [-1] }> { dims: &[len] };

candle-nn/src/moe/cutile.rs

@@ -16,7 +16,6 @@ const MAX_EXPERTS: usize = 1023;
 mod kernels {
     use candle::cutile;
     use cutile::core::*;
-    use cutile::cutile_compiler;
 
     #[cutile::entry(
         unchecked_accesses = true,