Kernels added

Cargo.toml

@@ -4,7 +4,16 @@ version = "0.1.0"
 edition = "2021"
 
 [dependencies]
+clap = { version = "4.5.13", features = ["derive"] }
+num_cpus = "1.16.0"
+rayon = "1.10.0"
 
 [[bin]]
 name = "bench"
-path = "src/main.rs"
+path = "src/main.rs"
+
+[profile.release]
+debug = false
+lto = true
+# codegen-units = 1
+incremental = false

Makefile

@@ -4,7 +4,7 @@ GIT_HASH := $(shell git rev-parse --short HEAD || echo 'development')
 CURRENT_TIME = $(shell date +"%Y-%m-%d:T%H:%M:%S")
 LD_FLAGS = '-s -X main.date=${CURRENT_TIME} -X main.version=${VERSION} -X main.commit=${GIT_HASH}'
 
-.PHONY: clean test check $(TARGET)
+.PHONY: clean test check install asm $(TARGET)
 
 .NOTPARALLEL:
 
@@ -17,12 +17,18 @@ $(TARGET):
 # 	$(info ===>  GENERATE swagger)
 # 	@go run github.com/swaggo/swag/cmd/swag init -d ./internal/api,./internal/util -g api.go -o ./api
 # 	@mv ./api/docs.go ./internal/api/docs.go
+install:
+	$(info ===>  INSTALL)
+	@cargo install cargo-asm
+
+asm:
+	$(info ===>  ASM)
+	@cargo rustc --release -- --emit asm
 
 clean:
 	$(info ===>  CLEAN)
 	@cargo clean
 	@rm -f $(TARGET)
-	@rm perf.data perf.data.old flamegraph.svg
 
 test:
 	$(info ===>  TESTING)

src/kernels/copy.rs

@@ -0,0 +1,12 @@
+use std::time::Instant;
+
+#[allow(clippy::ptr_arg)]
+pub fn copy(c: &mut Vec<f64>, a: &Vec<f64>, n: usize) -> f64 {
+    let s = Instant::now();
+
+    for i in 0..n {
+        c[i] = a[i];
+    }
+
+    s.elapsed().as_secs_f64()
+}

src/kernels/daxpy.rs

@@ -0,0 +1,12 @@
+use std::time::Instant;
+
+#[allow(clippy::ptr_arg)]
+pub fn daxpy(a: &mut Vec<f64>, b: &mut Vec<f64>, scalar: f64, n: usize) -> f64 {
+    let s = Instant::now();
+
+    for i in 0..n {
+        a[i] += scalar * b[i];
+    }
+
+    s.elapsed().as_secs_f64()
+}

src/kernels/init.rs

@@ -0,0 +1,12 @@
+use std::time::Instant;
+
+#[allow(clippy::ptr_arg)]
+pub fn init(b: &mut Vec<f64>, scalar: f64, n: usize) -> f64 {
+    let s = Instant::now();
+
+    for i in b.iter_mut().take(n) {
+        *i = scalar;
+    }
+
+    s.elapsed().as_secs_f64()
+}