Autogenerate entire prime butterfly files, rather than just chunks (#137

)

* Created a rust tool to autogenerate the full SSE prime butterfly file

* Removed unnecessary include

* added prime-specific benches

* formatting fix

* Make template architecture-agnostic

* Added wasm simd prime butterfly benches

* wasm simd refactor: rename the non-newtype util methods to _v128

* Hook up wasm simd to use the prime butterfly generator

* add neon prime butterfly benchmarks

* Check in cargo.lock for gen_simd_butterflies

* Converted neon prime butterflies over to the autogenerated system and FMA

* Use the same testing feature detection as neon

* Delete gen_sse_butterflies.py

* Exclude wasm simd benchmarks from rustfmt

* Simplify the template by pre-packaging the struct names

* Simplified butterfly macro invocation by removing unused parameters

* Fix macro parameter

* Fix struct name

* Use handlebars templates to avoid preprocessing

* Added "--check" mode to the simd butterfly script

* Added a CI step to verify autogenerated code match

* Test new CI step by submitting mismatching code

* Back out test code, remove handlebars import

.github/workflows/run_test.yml

@@ -46,6 +46,23 @@ jobs:
       - name: Run cargo fmt
         run: cargo fmt -- --check
 
+  autogeneration:
+    name: Check Autogenerated Code Match
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout sources
+        uses: actions/checkout@v3
+
+      - name: Install toolchain
+        uses: dtolnay/rust-toolchain@nightly
+
+      - name: Check SSE Prime Butterflies
+        run: cargo run --manifest-path ./tools/gen_simd_butterflies/Cargo.toml -- sse 7 11 13 17 19 23 29 31 --check src/sse/sse_prime_butterflies.rs
+      - name: Check NEON Prime Butterflies
+        run: cargo run --manifest-path ./tools/gen_simd_butterflies/Cargo.toml -- neon 7 11 13 17 19 23 29 31 --check src/neon/neon_prime_butterflies.rs
+      - name: Check Wasm SIMD Prime Butterflies
+        run: cargo run --manifest-path ./tools/gen_simd_butterflies/Cargo.toml -- wasm_simd 7 11 13 17 19 23 29 31 --check src/wasm_simd/wasm_simd_prime_butterflies.rs
+
   check_no_features:
     name: Check+Test no features
     runs-on: ubuntu-latest

.gitignore

@@ -1,4 +1,5 @@
 /target
 /Cargo.lock
+Tools/gen_simd_butterflies/target
 *.swp
 .vscode/

benches/bench_rustfft_neon.rs

@@ -80,6 +80,25 @@ fn bench_planned_multi_f64(b: &mut Bencher, len: usize) {
 #[bench] fn neon_butterfly32_31(b: &mut Bencher) { bench_planned_multi_f32(b, 31);}
 #[bench] fn neon_butterfly32_32(b: &mut Bencher) { bench_planned_multi_f32(b, 32);}
 
+
+#[bench] fn neon_prime_butterfly32_07(b: &mut Bencher) { bench_planned_multi_f32(b, 7);}
+#[bench] fn neon_prime_butterfly32_11(b: &mut Bencher) { bench_planned_multi_f32(b, 11);}
+#[bench] fn neon_prime_butterfly32_13(b: &mut Bencher) { bench_planned_multi_f32(b, 13);}
+#[bench] fn neon_prime_butterfly32_17(b: &mut Bencher) { bench_planned_multi_f32(b, 17);}
+#[bench] fn neon_prime_butterfly32_19(b: &mut Bencher) { bench_planned_multi_f32(b, 19);}
+#[bench] fn neon_prime_butterfly32_23(b: &mut Bencher) { bench_planned_multi_f32(b, 23);}
+#[bench] fn neon_prime_butterfly32_29(b: &mut Bencher) { bench_planned_multi_f32(b, 29);}
+#[bench] fn neon_prime_butterfly32_31(b: &mut Bencher) { bench_planned_multi_f32(b, 31);}
+
+#[bench] fn neon_prime_butterfly64_07(b: &mut Bencher) { bench_planned_multi_f64(b, 7);}
+#[bench] fn neon_prime_butterfly64_11(b: &mut Bencher) { bench_planned_multi_f64(b, 11);}
+#[bench] fn neon_prime_butterfly64_13(b: &mut Bencher) { bench_planned_multi_f64(b, 13);}
+#[bench] fn neon_prime_butterfly64_17(b: &mut Bencher) { bench_planned_multi_f64(b, 17);}
+#[bench] fn neon_prime_butterfly64_19(b: &mut Bencher) { bench_planned_multi_f64(b, 19);}
+#[bench] fn neon_prime_butterfly64_23(b: &mut Bencher) { bench_planned_multi_f64(b, 23);}
+#[bench] fn neon_prime_butterfly64_29(b: &mut Bencher) { bench_planned_multi_f64(b, 29);}
+#[bench] fn neon_prime_butterfly64_31(b: &mut Bencher) { bench_planned_multi_f64(b, 31);}
+
 #[bench] fn neon_butterfly64_02(b: &mut Bencher) { bench_planned_multi_f64(b, 2);}
 #[bench] fn neon_butterfly64_03(b: &mut Bencher) { bench_planned_multi_f64(b, 3);}
 #[bench] fn neon_butterfly64_04(b: &mut Bencher) { bench_planned_multi_f64(b, 4);}

benches/bench_rustfft_sse.rs

@@ -103,6 +103,24 @@ fn bench_planned_multi_f64(b: &mut Bencher, len: usize) {
 #[bench] fn sse_butterfly64_31(b: &mut Bencher) { bench_planned_multi_f64(b, 31);}
 #[bench] fn sse_butterfly64_32(b: &mut Bencher) { bench_planned_multi_f64(b, 32);}
 
+// prime butterflies
+#[bench] fn sse_prime_butterfly32_07(b: &mut Bencher) { bench_planned_multi_f32(b, 7);}
+#[bench] fn sse_prime_butterfly32_11(b: &mut Bencher) { bench_planned_multi_f32(b, 11);}
+#[bench] fn sse_prime_butterfly32_13(b: &mut Bencher) { bench_planned_multi_f32(b, 13);}
+#[bench] fn sse_prime_butterfly32_17(b: &mut Bencher) { bench_planned_multi_f32(b, 17);}
+#[bench] fn sse_prime_butterfly32_19(b: &mut Bencher) { bench_planned_multi_f32(b, 19);}
+#[bench] fn sse_prime_butterfly32_23(b: &mut Bencher) { bench_planned_multi_f32(b, 23);}
+#[bench] fn sse_prime_butterfly32_29(b: &mut Bencher) { bench_planned_multi_f32(b, 29);}
+#[bench] fn sse_prime_butterfly32_31(b: &mut Bencher) { bench_planned_multi_f32(b, 31);}
+#[bench] fn sse_prime_butterfly64_07(b: &mut Bencher) { bench_planned_multi_f64(b, 7);}
+#[bench] fn sse_prime_butterfly64_11(b: &mut Bencher) { bench_planned_multi_f64(b, 11);}
+#[bench] fn sse_prime_butterfly64_13(b: &mut Bencher) { bench_planned_multi_f64(b, 13);}
+#[bench] fn sse_prime_butterfly64_17(b: &mut Bencher) { bench_planned_multi_f64(b, 17);}
+#[bench] fn sse_prime_butterfly64_19(b: &mut Bencher) { bench_planned_multi_f64(b, 19);}
+#[bench] fn sse_prime_butterfly64_23(b: &mut Bencher) { bench_planned_multi_f64(b, 23);}
+#[bench] fn sse_prime_butterfly64_29(b: &mut Bencher) { bench_planned_multi_f64(b, 29);}
+#[bench] fn sse_prime_butterfly64_31(b: &mut Bencher) { bench_planned_multi_f64(b, 31);}
+
 // Powers of 2
 #[bench] fn sse_planned32_p2_00000064(b: &mut Bencher) { bench_planned_f32(b, 64); }
 #[bench] fn sse_planned32_p2_00000128(b: &mut Bencher) { bench_planned_f32(b, 128); }

benches/bench_rustfft_wasm_simd.rs

@@ -258,6 +258,24 @@ fn wasm_simd_butterfly64_32(b: &mut Bencher) {
     bench_planned_multi_f64(b, 32);
 }
 
+// prime butterflies
+#[bench] fn wasm_simd_prime_butterfly32_07(b: &mut Bencher) { bench_planned_multi_f32(b, 7);}
+#[bench] fn wasm_simd_prime_butterfly32_11(b: &mut Bencher) { bench_planned_multi_f32(b, 11);}
+#[bench] fn wasm_simd_prime_butterfly32_13(b: &mut Bencher) { bench_planned_multi_f32(b, 13);}
+#[bench] fn wasm_simd_prime_butterfly32_17(b: &mut Bencher) { bench_planned_multi_f32(b, 17);}
+#[bench] fn wasm_simd_prime_butterfly32_19(b: &mut Bencher) { bench_planned_multi_f32(b, 19);}
+#[bench] fn wasm_simd_prime_butterfly32_23(b: &mut Bencher) { bench_planned_multi_f32(b, 23);}
+#[bench] fn wasm_simd_prime_butterfly32_29(b: &mut Bencher) { bench_planned_multi_f32(b, 29);}
+#[bench] fn wasm_simd_prime_butterfly32_31(b: &mut Bencher) { bench_planned_multi_f32(b, 31);}
+#[bench] fn wasm_simd_prime_butterfly64_07(b: &mut Bencher) { bench_planned_multi_f64(b, 7);}
+#[bench] fn wasm_simd_prime_butterfly64_11(b: &mut Bencher) { bench_planned_multi_f64(b, 11);}
+#[bench] fn wasm_simd_prime_butterfly64_13(b: &mut Bencher) { bench_planned_multi_f64(b, 13);}
+#[bench] fn wasm_simd_prime_butterfly64_17(b: &mut Bencher) { bench_planned_multi_f64(b, 17);}
+#[bench] fn wasm_simd_prime_butterfly64_19(b: &mut Bencher) { bench_planned_multi_f64(b, 19);}
+#[bench] fn wasm_simd_prime_butterfly64_23(b: &mut Bencher) { bench_planned_multi_f64(b, 23);}
+#[bench] fn wasm_simd_prime_butterfly64_29(b: &mut Bencher) { bench_planned_multi_f64(b, 29);}
+#[bench] fn wasm_simd_prime_butterfly64_31(b: &mut Bencher) { bench_planned_multi_f64(b, 31);}
+
 // Powers of 2
 #[bench]
 fn wasm_simd_planned32_p2_00000064(b: &mut Bencher) {

rustfmt.toml

@@ -4,8 +4,12 @@ ignore = [
     "benches/bench_rustfft.rs", 
     "benches/bench_rustfft_sse.rs",
     "benches/bench_rustfft_neon.rs", 
+    "benches/bench_rustfft_wasm_simd.rs", 
+    "benches/bench_compare_scalar_neon.rs",
+
+    # disable the prime butterfly files because they're autogenerated
     "src/sse/sse_prime_butterflies.rs", 
     "src/neon/neon_prime_butterflies.rs",
-    "benches/bench_compare_scalar_neon.rs",
-    ]
+    "src/wasm_simd/wasm_simd_prime_butterflies.rs",
+]