Rust-GPU · LegNeato · Nov 21, 2024 · Nov 21, 2024
diff --git a/blog/2024-11-21-optimizing-matrix-mul/code/Cargo.lock b/blog/2024-11-21-optimizing-matrix-mul/code/Cargo.lock
diff --git a/blog/2024-11-21-optimizing-matrix-mul/code/Cargo.toml b/blog/2024-11-21-optimizing-matrix-mul/code/Cargo.toml
@@ -9,7 +9,7 @@ members = [
     "crates/gpu/workgroup_2d",
     "crates/gpu/tiling_1d",
     "crates/gpu/tiling_1d_loop",
-    "crates/gpu/tiling_2d_simd",
+    "crates/gpu/tiling_2d",
     #
     # ---- The rust code that runs both on the GPU and the CPU. ----
     # It "knows" what platform it is being compiled for and can conditionally change
@@ -31,7 +31,8 @@ members = [
     "crates/cpu/compiled_for_gpu/workgroup_256",
     "crates/cpu/compiled_for_gpu/workgroup_2d",
     "crates/cpu/compiled_for_gpu/tiling_1d",
-    "crates/cpu/compiled_for_gpu/tiling_2d_simd",
+    "crates/cpu/compiled_for_gpu/tiling_1d_loop",
+    "crates/cpu/compiled_for_gpu/tiling_2d",
     "crates/cpu/compiled_for_gpu/isomorphic",
     # 3) A binary that runs on the CPU. It configures the `matmul` library on the CPU
     #    and then tells it to run the matrix multiplication.

diff --git a/blog/2024-11-21-optimizing-matrix-mul/code/benches/gpu_bench.rs b/blog/2024-11-21-optimizing-matrix-mul/code/benches/gpu_bench.rs
@@ -38,7 +38,7 @@ fn bench_all_variants(c: &mut Criterion) {
     let multiplier_workgroup_2d = matmul::workgroup_2d::wgpu();
     let multiplier_tiling_1d = matmul::tiling_1d::wgpu();
     let multiplier_tiling_1d_loop = matmul::tiling_1d_loop::wgpu();
-    let multiplier_tiling_2d_simd = matmul::tiling_2d_simd::wgpu();
+    let multiplier_tiling_2d = matmul::tiling_2d::wgpu();
     let multiplier_isomorphic_gpu = matmul::isomorphic::wgpu();
 
     for &(m, k, n) in SIZES {
@@ -126,17 +126,11 @@ fn bench_all_variants(c: &mut Criterion) {
         );
 
         group.bench_with_input(
-            BenchmarkId::new("tiling_2d_simd:wgpu", format!("{}x{}x{}", m, k, n)),
+            BenchmarkId::new("tiling_2d:wgpu", format!("{}x{}x{}", m, k, n)),
             &(m, k, n),
             |bench, &(m, k, n)| {
                 bench.iter(|| {
-                    black_box(multiplier_tiling_2d_simd.multiply(
-                        black_box(&a),
-                        black_box(&b),
-                        m,
-                        k,
-                        n,
-                    ))
+                    black_box(multiplier_tiling_2d.multiply(black_box(&a), black_box(&b), m, k, n))
                 });
             },
         );

diff --git a/blog/2024-11-21-optimizing-matrix-mul/code/bin/blog/src/bin.rs b/blog/2024-11-21-optimizing-matrix-mul/code/bin/blog/src/bin.rs
@@ -32,7 +32,7 @@ fn main() {
     run_tests(matmul::workgroup_2d::wgpu(), &sizes);
     run_tests(matmul::tiling_1d::wgpu(), &sizes);
     run_tests(matmul::tiling_1d_loop::wgpu(), &sizes);
-    run_tests(matmul::tiling_2d_simd::wgpu(), &sizes);
+    run_tests(matmul::tiling_2d::wgpu(), &sizes);
 
     run_tests(matmul::isomorphic::wgpu(), &sizes);
     run_tests(matmul::isomorphic::cpu::single_threaded(), &sizes);

diff --git a/blog/2024-11-21-optimizing-matrix-mul/code/crates/cpu/compiled_for_gpu/isomorphic/build.rs b/blog/2024-11-21-optimizing-matrix-mul/code/crates/cpu/compiled_for_gpu/isomorphic/build.rs
@@ -4,7 +4,7 @@ use std::fs;
 use std::path::{Path, PathBuf};
 
 fn main() -> Result<(), Box<dyn std::error::Error>> {
-    let gpu_crate_path = Path::new("../../../gpu/tiling_2d_simd");
+    let gpu_crate_path = Path::new("../../../gpu/tiling_2d");
 
     // Compile the shader crate with SpirvBuilder.
     let result = SpirvBuilder::new(gpu_crate_path, "spirv-unknown-vulkan1.2")

diff --git a/...ompiled_for_gpu/tiling_2d_simd/Cargo.toml → ...cpu/compiled_for_gpu/tiling_2d/Cargo.toml b/...ompiled_for_gpu/tiling_2d_simd/Cargo.toml → ...cpu/compiled_for_gpu/tiling_2d/Cargo.toml
@@ -1,5 +1,5 @@
 [package]
-name = "compiled_tiling_2d_simd"
+name = "compiled_tiling_2d"
 version = "0.1.0"
 edition = "2021"
 

diff --git a/.../compiled_for_gpu/tiling_2d_simd/build.rs → ...s/cpu/compiled_for_gpu/tiling_2d/build.rs b/.../compiled_for_gpu/tiling_2d_simd/build.rs → ...s/cpu/compiled_for_gpu/tiling_2d/build.rs
@@ -4,7 +4,7 @@ use std::fs;
 use std::path::{Path, PathBuf};
 
 fn main() -> Result<(), Box<dyn std::error::Error>> {
-    let gpu_crate_path = Path::new("../../../gpu/tiling_2d_simd");
+    let gpu_crate_path = Path::new("../../../gpu/tiling_2d");
 
     // Compile the shader crate with SpirvBuilder.
     let result = SpirvBuilder::new(gpu_crate_path, "spirv-unknown-vulkan1.2")

diff --git a/...ompiled_for_gpu/tiling_2d_simd/src/lib.rs → ...cpu/compiled_for_gpu/tiling_2d/src/lib.rs b/...ompiled_for_gpu/tiling_2d_simd/src/lib.rs → ...cpu/compiled_for_gpu/tiling_2d/src/lib.rs
diff --git a/blog/2024-11-21-optimizing-matrix-mul/code/crates/cpu/matmul/Cargo.toml b/blog/2024-11-21-optimizing-matrix-mul/code/crates/cpu/matmul/Cargo.toml
@@ -22,7 +22,7 @@ compiled_workgroup_256 = { path = "../compiled_for_gpu/workgroup_256" }
 compiled_workgroup_2d = { path = "../compiled_for_gpu/workgroup_2d" }
 compiled_tiling_1d = { path = "../compiled_for_gpu/tiling_1d" }
 compiled_tiling_1d_loop = { path = "../compiled_for_gpu/tiling_1d_loop" }
-compiled_tiling_2d_simd = { path = "../compiled_for_gpu/tiling_2d_simd" }
+compiled_tiling_2d = { path = "../compiled_for_gpu/tiling_2d" }
 compiled_isomorphic = { path = "../compiled_for_gpu/isomorphic" }
 # The CPU side of the isomophic implementation.
 isomorphic = { path = "../../shared/isomorphic" }
diff --git a/blog/2024-11-21-optimizing-matrix-mul/code/crates/cpu/matmul/src/lib.rs b/blog/2024-11-21-optimizing-matrix-mul/code/crates/cpu/matmul/src/lib.rs
@@ -86,7 +86,7 @@ pub mod tiling_1d_loop {
     }
 }
 
-pub mod tiling_2d_simd {
+pub mod tiling_2d {
     use super::*;
     use crate::backends::wgpu::MatrixMultiplier;
 

diff --git a/blog/2024-11-21-optimizing-matrix-mul/code/crates/cpu/matmul/src/variants.rs b/blog/2024-11-21-optimizing-matrix-mul/code/crates/cpu/matmul/src/variants.rs
@@ -158,13 +158,13 @@ pub struct Tiling2dSimd;
 
 impl Display for Tiling2dSimd {
     fn fmt(&self, f: &mut Formatter<'_>) -> fmt::Result {
-        write!(f, "tiling_2d_simd")
+        write!(f, "tiling_2d")
     }
 }
 
 impl Gpu for Tiling2dSimd {
     fn compiled_shader(&self) -> &[u8] {
-        compiled_tiling_2d_simd::SHADER_BINARY
+        compiled_tiling_2d::SHADER_BINARY
     }
 }
 

diff --git a/...code/crates/gpu/tiling_2d_simd/Cargo.lock → ...-mul/code/crates/gpu/tiling_2d/Cargo.lock b/...code/crates/gpu/tiling_2d_simd/Cargo.lock → ...-mul/code/crates/gpu/tiling_2d/Cargo.lock
diff --git a/...code/crates/gpu/tiling_2d_simd/Cargo.toml → ...-mul/code/crates/gpu/tiling_2d/Cargo.toml b/...code/crates/gpu/tiling_2d_simd/Cargo.toml → ...-mul/code/crates/gpu/tiling_2d/Cargo.toml
@@ -1,5 +1,5 @@
 [package]
-name = "tiling_2d_simd"
+name = "tiling_2d"
 version = "0.1.0"
 edition = "2021"
 

diff --git a/...code/crates/gpu/tiling_2d_simd/src/lib.rs → ...-mul/code/crates/gpu/tiling_2d/src/lib.rs b/...code/crates/gpu/tiling_2d_simd/src/lib.rs → ...-mul/code/crates/gpu/tiling_2d/src/lib.rs
diff --git a/blog/2024-11-21-optimizing-matrix-mul/index.md b/blog/2024-11-21-optimizing-matrix-mul/index.md
@@ -291,7 +291,7 @@ We can take this a step further and calculate 2D results per thread! Instead of
 calculating 4 elements per single row, we can calculate 4 elements for 4 rows (e.g. a 2D
 tile).
 
-import { RustTiling2dSimd } from './snippets/tiling_2d_simd.tsx';
+import { RustTiling2d } from './snippets/tiling_2d.tsx';
 
 <RustTiling2dSimd />
 

diff --git a/...ng-matrix-mul/snippets/tiling_2d_simd.tsx → ...imizing-matrix-mul/snippets/tiling_2d.tsx b/...ng-matrix-mul/snippets/tiling_2d_simd.tsx → ...imizing-matrix-mul/snippets/tiling_2d.tsx
@@ -1,6 +1,6 @@
 import React from "react";
 import Snippet from "@site/src/components/Snippet";
-import RustKernelSource from "!!raw-loader!../code/crates/gpu/tiling_2d_simd/src/lib.rs";
+import RustKernelSource from "!!raw-loader!../code/crates/gpu/tiling_2d/src/lib.rs";
 
 export const RustTiling2dSimd: React.FC = () => (
     <Snippet language="rust" className="text-xs" title="2D tiling kernel with Rust GPU">