diff --git a/hercules_samples/call/build.rs b/hercules_samples/call/build.rs
index af48fe64f7c2b778c9841e45ecf13b8a6a5740d2..7f5816ce0992e02e0c9d4da83617d8b8414eb3c2 100644
--- a/hercules_samples/call/build.rs
+++ b/hercules_samples/call/build.rs
@@ -4,6 +4,8 @@ fn main() {
JunoCompiler::new()
.ir_in_src("call.hir")
.unwrap()
+ .schedule_in_src(if cfg!(feature = "cuda") { "gpu.sch" } else { "cpu.sch" })
+ .unwrap()
.build()
.unwrap();
}
diff --git a/hercules_samples/call/src/cpu.sch b/hercules_samples/call/src/cpu.sch
new file mode 100644
index 0000000000000000000000000000000000000000..4c684da2f176dd3675dc79fa506d2d60fe6e0577
--- /dev/null
+++ b/hercules_samples/call/src/cpu.sch
@@ -0,0 +1,19 @@
+gvn(*);
+phi-elim(*);
+dce(*);
+
+auto-outline(*);
+
+ip-sroa(*);
+sroa(*);
+fork-split(*);
+unforkify(*);
+dce(*);
+float-collections(*);
+gvn(*);
+phi-elim(*);
+dce(*);
+
+infer-schedules(*);
+
+gcm(*);
diff --git a/hercules_samples/call/src/gpu.sch b/hercules_samples/call/src/gpu.sch
new file mode 100644
index 0000000000000000000000000000000000000000..1e654e22664a5b9c79a20888d1f013f9256d7d82
--- /dev/null
+++ b/hercules_samples/call/src/gpu.sch
@@ -0,0 +1,18 @@
+gvn(*);
+phi-elim(*);
+dce(*);
+
+let out = auto-outline(*);
+gpu(out.add);
+
+ip-sroa(*);
+sroa(*);
+dce(*);
+float-collections(*);
+gvn(*);
+phi-elim(*);
+dce(*);
+
+infer-schedules(*);
+
+gcm(*);
diff --git a/hercules_samples/ccp/build.rs b/hercules_samples/ccp/build.rs
index f04d48c7d0ea6df8b16d70b05cedabfc04c1f6f2..c98d0551654113ee26523179feae3322383f0d01 100644
--- a/hercules_samples/ccp/build.rs
+++ b/hercules_samples/ccp/build.rs
@@ -4,6 +4,8 @@ fn main() {
JunoCompiler::new()
.ir_in_src("ccp.hir")
.unwrap()
+ .schedule_in_src(if cfg!(feature = "cuda") { "gpu.sch" } else { "cpu.sch" })
+ .unwrap()
.build()
.unwrap();
}
diff --git a/hercules_samples/ccp/src/cpu.sch b/hercules_samples/ccp/src/cpu.sch
new file mode 100644
index 0000000000000000000000000000000000000000..4c684da2f176dd3675dc79fa506d2d60fe6e0577
--- /dev/null
+++ b/hercules_samples/ccp/src/cpu.sch
@@ -0,0 +1,19 @@
+gvn(*);
+phi-elim(*);
+dce(*);
+
+auto-outline(*);
+
+ip-sroa(*);
+sroa(*);
+fork-split(*);
+unforkify(*);
+dce(*);
+float-collections(*);
+gvn(*);
+phi-elim(*);
+dce(*);
+
+infer-schedules(*);
+
+gcm(*);
diff --git a/hercules_samples/ccp/src/gpu.sch b/hercules_samples/ccp/src/gpu.sch
new file mode 100644
index 0000000000000000000000000000000000000000..d8f6a2d00171255a1596d73dc9bf4ecff7ff308a
--- /dev/null
+++ b/hercules_samples/ccp/src/gpu.sch
@@ -0,0 +1,18 @@
+gvn(*);
+phi-elim(*);
+dce(*);
+
+let out = auto-outline(*);
+gpu(out.tricky);
+
+ip-sroa(*);
+sroa(*);
+dce(*);
+float-collections(*);
+gvn(*);
+phi-elim(*);
+dce(*);
+
+infer-schedules(*);
+
+gcm(*);
diff --git a/hercules_samples/dot/src/gpu.sch b/hercules_samples/dot/src/gpu.sch
index a1a5108882971f4cc0baabb4646a19002beba0ad..4adbf530a7ce8277cd8c691349f36de65ba4e251 100644
--- a/hercules_samples/dot/src/gpu.sch
+++ b/hercules_samples/dot/src/gpu.sch
@@ -2,9 +2,8 @@ gvn(*);
phi-elim(*);
dce(*);
-auto-outline(*);
-gpu(*);
-host(dot);
+let out = auto-outline(*);
+gpu(out.dot);
ip-sroa(*);
sroa(*);
diff --git a/hercules_samples/fac/build.rs b/hercules_samples/fac/build.rs
index 4d8226f11183d9500e6affec4c46110e8626ee69..1986a74679dc43c19ec4638f7f1d96ef8832d769 100644
--- a/hercules_samples/fac/build.rs
+++ b/hercules_samples/fac/build.rs
@@ -4,6 +4,8 @@ fn main() {
JunoCompiler::new()
.ir_in_src("fac.hir")
.unwrap()
+ .schedule_in_src(if cfg!(feature = "cuda") { "gpu.sch" } else { "cpu.sch" })
+ .unwrap()
.build()
.unwrap();
}
diff --git a/hercules_samples/fac/src/cpu.sch b/hercules_samples/fac/src/cpu.sch
new file mode 100644
index 0000000000000000000000000000000000000000..4c684da2f176dd3675dc79fa506d2d60fe6e0577
--- /dev/null
+++ b/hercules_samples/fac/src/cpu.sch
@@ -0,0 +1,19 @@
+gvn(*);
+phi-elim(*);
+dce(*);
+
+auto-outline(*);
+
+ip-sroa(*);
+sroa(*);
+fork-split(*);
+unforkify(*);
+dce(*);
+float-collections(*);
+gvn(*);
+phi-elim(*);
+dce(*);
+
+infer-schedules(*);
+
+gcm(*);
diff --git a/hercules_samples/fac/src/gpu.sch b/hercules_samples/fac/src/gpu.sch
new file mode 100644
index 0000000000000000000000000000000000000000..1885854ca2d2b05e000d64165244080060b2a4f9
--- /dev/null
+++ b/hercules_samples/fac/src/gpu.sch
@@ -0,0 +1,18 @@
+gvn(*);
+phi-elim(*);
+dce(*);
+
+let out = auto-outline(*);
+gpu(out.fac);
+
+ip-sroa(*);
+sroa(*);
+dce(*);
+float-collections(*);
+gvn(*);
+phi-elim(*);
+dce(*);
+
+infer-schedules(*);
+
+gcm(*);
diff --git a/hercules_samples/matmul/src/gpu.sch b/hercules_samples/matmul/src/gpu.sch
index c9d6b3361f7f48bc47f0d6ee2da3fc9018cebf48..9a714789ac768405a2782cac6e68338c2f0697bb 100644
--- a/hercules_samples/matmul/src/gpu.sch
+++ b/hercules_samples/matmul/src/gpu.sch
@@ -2,9 +2,8 @@ gvn(*);
phi-elim(*);
dce(*);
-auto-outline(*);
-gpu(*);
-host(matmul);
+let out = auto-outline(*);
+gpu(out.matmul);
ip-sroa(*);
sroa(*);
diff --git a/hercules_samples/matmul/src/main.rs b/hercules_samples/matmul/src/main.rs
index 7b6cfe79048df5f6ff22a83b4e5859b2338f2b9d..abd25ec9cddbd4be508b3f484cffc1df1365dc4d 100644
--- a/hercules_samples/matmul/src/main.rs
+++ b/hercules_samples/matmul/src/main.rs
@@ -33,15 +33,13 @@ fn main() {
}
#[cfg(feature = "cuda")]
{
- let a_box = CUDABox::from_cpu_ref(HerculesCPURef::from_slice(&mut a));
- let a = a_box.get_ref();
- let b_box = CUDABox::from_cpu_ref(HerculesCPURef::from_slice(&mut b));
- let b = b_box.get_ref();
+ let a = CUDABox::from_cpu_ref(HerculesCPURef::from_slice(&mut a));
+ let b = CUDABox::from_cpu_ref(HerculesCPURef::from_slice(&mut b));
let mut r = runner!(matmul);
- let c = r.run(I as u64, J as u64, K as u64, a, b).await;
+ let c = r.run(I as u64, J as u64, K as u64, a.get_ref(), b.get_ref()).await;
let mut c_cpu: Box<[i32]> = vec![0; correct_c.len()].into_boxed_slice();
c.to_cpu_ref(&mut c_cpu);
- assert_eq!(c_cpu.as_ref(), &*correct_c);
+ assert_eq!(&*c_cpu, &*correct_c);
}
});
}
diff --git a/juno_samples/antideps/build.rs b/juno_samples/antideps/build.rs
index 7ed716a444460d7a90965f5b7f5faf3a7aadcb14..92b30c43b385812c02f2685ab3cdc76950dd6960 100644
--- a/juno_samples/antideps/build.rs
+++ b/juno_samples/antideps/build.rs
@@ -1,6 +1,15 @@
use juno_build::JunoCompiler;
fn main() {
+ #[cfg(feature = "cuda")]
+ JunoCompiler::new()
+ .file_in_src("antideps.jn")
+ .unwrap()
+ .schedule_in_src("gpu.sch")
+ .unwrap()
+ .build()
+ .unwrap();
+ #[cfg(not(feature = "cuda"))]
JunoCompiler::new()
.file_in_src("antideps.jn")
.unwrap()
diff --git a/juno_samples/antideps/src/gpu.sch b/juno_samples/antideps/src/gpu.sch
new file mode 100644
index 0000000000000000000000000000000000000000..d3f4a6c28ccefa44b2dd474e7f54546f1a21eab6
--- /dev/null
+++ b/juno_samples/antideps/src/gpu.sch
@@ -0,0 +1,18 @@
+gvn(*);
+phi-elim(*);
+dce(*);
+
+let out = auto-outline(*);
+gpu(out.simple_antideps, out.loop_antideps, out.complex_antideps1, out.complex_antideps2, out.very_complex_antideps, out.read_chains, out.array_of_structs);
+
+ip-sroa(*);
+sroa(*);
+dce(*);
+float-collections(*);
+gvn(*);
+phi-elim(*);
+dce(*);
+
+infer-schedules(*);
+
+gcm(*);
diff --git a/juno_samples/casts_and_intrinsics/build.rs b/juno_samples/casts_and_intrinsics/build.rs
index 16d5c7a4f7fcb00344fc7669b67103a27f71a7c6..e43a2ac82f2dc60dc1198a9884cce617aa3d70f5 100644
--- a/juno_samples/casts_and_intrinsics/build.rs
+++ b/juno_samples/casts_and_intrinsics/build.rs
@@ -1,6 +1,15 @@
use juno_build::JunoCompiler;
fn main() {
+ #[cfg(feature = "cuda")]
+ JunoCompiler::new()
+ .file_in_src("casts_and_intrinsics.jn")
+ .unwrap()
+ .schedule_in_src("gpu.sch")
+ .unwrap()
+ .build()
+ .unwrap();
+ #[cfg(not(feature = "cuda"))]
JunoCompiler::new()
.file_in_src("casts_and_intrinsics.jn")
.unwrap()
diff --git a/juno_samples/casts_and_intrinsics/src/gpu.sch b/juno_samples/casts_and_intrinsics/src/gpu.sch
new file mode 100644
index 0000000000000000000000000000000000000000..b2fb3449954ca567ef3b24adebcf5610763a687e
--- /dev/null
+++ b/juno_samples/casts_and_intrinsics/src/gpu.sch
@@ -0,0 +1,18 @@
+gvn(*);
+phi-elim(*);
+dce(*);
+
+let out = auto-outline(*);
+gpu(out.casts_and_intrinsics);
+
+ip-sroa(*);
+sroa(*);
+dce(*);
+float-collections(*);
+gvn(*);
+phi-elim(*);
+dce(*);
+
+infer-schedules(*);
+
+gcm(*);
diff --git a/juno_samples/cava/build.rs b/juno_samples/cava/build.rs
index 929d3eba3e1c83f185c2c0ff256450b05247c80d..03d54160d9a10d9cf43f9b3f057d45696e6e6a22 100644
--- a/juno_samples/cava/build.rs
+++ b/juno_samples/cava/build.rs
@@ -2,6 +2,15 @@ extern crate juno_build;
use juno_build::JunoCompiler;
fn main() {
+ #[cfg(feature = "cuda")]
+ JunoCompiler::new()
+ .file_in_src("cava.jn")
+ .unwrap()
+ .schedule_in_src("gpu.sch")
+ .unwrap()
+ .build()
+ .unwrap();
+ #[cfg(not(feature = "cuda"))]
JunoCompiler::new()
.file_in_src("cava.jn")
.unwrap()
diff --git a/juno_samples/cava/src/gpu.sch b/juno_samples/cava/src/gpu.sch
new file mode 100644
index 0000000000000000000000000000000000000000..07f71c99302c86a9ec3dc5c813b854fe3af983b0
--- /dev/null
+++ b/juno_samples/cava/src/gpu.sch
@@ -0,0 +1,18 @@
+gvn(*);
+phi-elim(*);
+dce(*);
+
+gpu(scale, demosaic, medianMatrix, transform, gamut, tone_map, descale);
+
+ip-sroa(*);
+sroa(*);
+
+dce(*);
+float-collections(*);
+gvn(*);
+phi-elim(*);
+dce(*);
+
+infer-schedules(*);
+
+gcm(*);
diff --git a/juno_samples/concat/build.rs b/juno_samples/concat/build.rs
index f7784b999492955289fc83ad1297907c2d0ce996..c91df94e2fbd0c05226cedb3f437e22c4689a961 100644
--- a/juno_samples/concat/build.rs
+++ b/juno_samples/concat/build.rs
@@ -4,6 +4,8 @@ fn main() {
JunoCompiler::new()
.file_in_src("concat.jn")
.unwrap()
+ .schedule_in_src(if cfg!(feature = "cuda") { "gpu.sch" } else { "cpu.sch" })
+ .unwrap()
.build()
.unwrap();
}
diff --git a/juno_samples/concat/src/concat.jn b/juno_samples/concat/src/concat.jn
index 2471671e69af7e9c73e2347abfe56e2db722d1d0..b9806c9348e73a3467f29f2f39c981b8f347037c 100644
--- a/juno_samples/concat/src/concat.jn
+++ b/juno_samples/concat/src/concat.jn
@@ -18,15 +18,7 @@ fn sum<t : number, c : usize>(arr : t[c]) -> t {
}
#[entry]
-fn concat_entry(a : i32) -> i32 {
- let arr1 : i32[3];
- let arr2 : i32[6];
- arr1[0] = a;
- arr1[1] = a;
- arr2[0] = a;
- arr2[1] = a;
- arr2[4] = a;
- arr2[5] = a;
+fn concat_entry(arr1 : i32[3], arr2 : i32[6]) -> i32 {
let arr3 = concat::<i32, 3, 6>(arr1, arr2);
return sum::<i32, 9>(arr3);
}
diff --git a/juno_samples/concat/src/cpu.sch b/juno_samples/concat/src/cpu.sch
new file mode 100644
index 0000000000000000000000000000000000000000..680adaeb2070b26b68fba361aafb64e5af2afc03
--- /dev/null
+++ b/juno_samples/concat/src/cpu.sch
@@ -0,0 +1,17 @@
+gvn(*);
+phi-elim(*);
+dce(*);
+
+cpu(concat, sum);
+
+ip-sroa(*);
+sroa(*);
+dce(*);
+float-collections(*);
+gvn(*);
+phi-elim(*);
+dce(*);
+
+infer-schedules(*);
+
+gcm(*);
diff --git a/juno_samples/concat/src/gpu.sch b/juno_samples/concat/src/gpu.sch
new file mode 100644
index 0000000000000000000000000000000000000000..8ee4ef0e5627aa6a51cb5e86dccd014f18dcfd96
--- /dev/null
+++ b/juno_samples/concat/src/gpu.sch
@@ -0,0 +1,17 @@
+gvn(*);
+phi-elim(*);
+dce(*);
+
+gpu(concat, sum);
+
+ip-sroa(*);
+sroa(*);
+dce(*);
+float-collections(*);
+gvn(*);
+phi-elim(*);
+dce(*);
+
+infer-schedules(*);
+
+gcm(*);
diff --git a/juno_samples/concat/src/main.rs b/juno_samples/concat/src/main.rs
index db3f37fdaa6047d146ebd899cc7178e2b135d7ee..d0929fbf0a3d5d5073b51f79ee8ab575208eb7cc 100644
--- a/juno_samples/concat/src/main.rs
+++ b/juno_samples/concat/src/main.rs
@@ -1,15 +1,35 @@
#![feature(concat_idents)]
use hercules_rt::runner;
+use hercules_rt::HerculesCPURef;
+#[cfg(feature = "cuda")]
+use hercules_rt::CUDABox;
juno_build::juno!("concat");
fn main() {
async_std::task::block_on(async {
let mut r = runner!(concat_entry);
- let output = r.run(7).await;
- println!("{}", output);
- assert_eq!(output, 42);
+ #[cfg(not(feature = "cuda"))]
+ {
+ let mut a_data = [7, 7, 0];
+ let a = HerculesCPURef::from_slice(&mut a_data);
+ let mut b_data = [7, 7, 0, 0, 7, 7];
+ let b = HerculesCPURef::from_slice(&mut b_data);
+ let output = r.run(a, b).await;
+ println!("{}", output);
+ assert_eq!(output, 42);
+ }
+ #[cfg(feature = "cuda")]
+ {
+ let mut a_data = [7, 7, 0];
+ let a = CUDABox::from_cpu_ref(HerculesCPURef::from_slice(&mut a_data));
+ let mut b_data = [7, 7, 0, 0, 7, 7];
+ let b = CUDABox::from_cpu_ref(HerculesCPURef::from_slice(&mut b_data));
+ let output = r.run(a.get_ref(), b.get_ref()).await;
+ println!("{}", output);
+ assert_eq!(output, 42);
+ }
});
}
diff --git a/juno_samples/cpu.sch b/juno_samples/cpu.sch
new file mode 100644
index 0000000000000000000000000000000000000000..4c684da2f176dd3675dc79fa506d2d60fe6e0577
--- /dev/null
+++ b/juno_samples/cpu.sch
@@ -0,0 +1,19 @@
+gvn(*);
+phi-elim(*);
+dce(*);
+
+auto-outline(*);
+
+ip-sroa(*);
+sroa(*);
+fork-split(*);
+unforkify(*);
+dce(*);
+float-collections(*);
+gvn(*);
+phi-elim(*);
+dce(*);
+
+infer-schedules(*);
+
+gcm(*);
diff --git a/juno_samples/gpu.sch b/juno_samples/gpu.sch
new file mode 100644
index 0000000000000000000000000000000000000000..9a714789ac768405a2782cac6e68338c2f0697bb
--- /dev/null
+++ b/juno_samples/gpu.sch
@@ -0,0 +1,18 @@
+gvn(*);
+phi-elim(*);
+dce(*);
+
+let out = auto-outline(*);
+gpu(out.matmul);
+
+ip-sroa(*);
+sroa(*);
+dce(*);
+float-collections(*);
+gvn(*);
+phi-elim(*);
+dce(*);
+
+infer-schedules(*);
+
+gcm(*);
diff --git a/juno_samples/implicit_clone/build.rs b/juno_samples/implicit_clone/build.rs
index 75c1afc41a75b2006b26042323df3bdc3fcf5a17..dc134e59757a301b2be13ab61ed4d2feb252d1fe 100644
--- a/juno_samples/implicit_clone/build.rs
+++ b/juno_samples/implicit_clone/build.rs
@@ -1,6 +1,15 @@
use juno_build::JunoCompiler;
fn main() {
+ #[cfg(feature = "cuda")]
+ JunoCompiler::new()
+ .file_in_src("implicit_clone.jn")
+ .unwrap()
+ .schedule_in_src("gpu.sch")
+ .unwrap()
+ .build()
+ .unwrap();
+ #[cfg(not(feature = "cuda"))]
JunoCompiler::new()
.file_in_src("implicit_clone.jn")
.unwrap()
diff --git a/juno_samples/implicit_clone/src/gpu.sch b/juno_samples/implicit_clone/src/gpu.sch
new file mode 100644
index 0000000000000000000000000000000000000000..443fc778fe6b6f4489d6c5a339b691f488f917c2
--- /dev/null
+++ b/juno_samples/implicit_clone/src/gpu.sch
@@ -0,0 +1,18 @@
+gvn(*);
+phi-elim(*);
+dce(*);
+
+let out = auto-outline(*);
+gpu(out.simple_implicit_clone, out.loop_implicit_clone, out.double_loop_implicit_clone, out.tricky_loop_implicit_clone, out.tricky2_loop_implicit_clone, out.tricky3_loop_implicit_clone, out.no_implicit_clone, out.mirage_implicit_clone);
+
+ip-sroa(*);
+sroa(*);
+dce(*);
+float-collections(*);
+gvn(*);
+phi-elim(*);
+dce(*);
+
+infer-schedules(*);
+
+gcm(*);
diff --git a/juno_samples/matmul/build.rs b/juno_samples/matmul/build.rs
index 926fbc33ecfa5ab31b40a92f778bb4d3b7f6a77e..ff3e3d8c79091349f9d76716f3d7496bba3d5503 100644
--- a/juno_samples/matmul/build.rs
+++ b/juno_samples/matmul/build.rs
@@ -1,6 +1,15 @@
use juno_build::JunoCompiler;
fn main() {
+ #[cfg(feature = "cuda")]
+ JunoCompiler::new()
+ .file_in_src("matmul.jn")
+ .unwrap()
+ .schedule_in_src("gpu.sch")
+ .unwrap()
+ .build()
+ .unwrap();
+ #[cfg(not(feature = "cuda"))]
JunoCompiler::new()
.file_in_src("matmul.jn")
.unwrap()
diff --git a/juno_samples/matmul/src/gpu.sch b/juno_samples/matmul/src/gpu.sch
new file mode 100644
index 0000000000000000000000000000000000000000..e85dafdfdbfac2021b0215f625fb8077972be6b4
--- /dev/null
+++ b/juno_samples/matmul/src/gpu.sch
@@ -0,0 +1,18 @@
+gvn(*);
+phi-elim(*);
+dce(*);
+
+let out = auto-outline(*);
+gpu(out.matmul, out.tiled_64_matmul);
+
+ip-sroa(*);
+sroa(*);
+dce(*);
+float-collections(*);
+gvn(*);
+phi-elim(*);
+dce(*);
+
+infer-schedules(*);
+
+gcm(*);
diff --git a/juno_samples/matmul/src/main.rs b/juno_samples/matmul/src/main.rs
index fa5d1f04d48cdf48cf377e8f3d08de80d30e688e..50fe1760eeeedc946d510a6d5285d76e1346f3cc 100644
--- a/juno_samples/matmul/src/main.rs
+++ b/juno_samples/matmul/src/main.rs
@@ -3,6 +3,8 @@
use rand::random;
use hercules_rt::{runner, HerculesCPURef};
+#[cfg(feature = "cuda")]
+use hercules_rt::CUDABox;
juno_build::juno!("matmul");
@@ -11,8 +13,8 @@ fn main() {
const I: usize = 256;
const J: usize = 64;
const K: usize = 128;
- let a: Box<[i32]> = (0..I * J).map(|_| random::<i32>() % 100).collect();
- let b: Box<[i32]> = (0..J * K).map(|_| random::<i32>() % 100).collect();
+ let mut a: Box<[i32]> = (0..I * J).map(|_| random::<i32>() % 100).collect();
+ let mut b: Box<[i32]> = (0..J * K).map(|_| random::<i32>() % 100).collect();
let mut correct_c: Box<[i32]> = (0..I * K).map(|_| 0).collect();
for i in 0..I {
for k in 0..K {
@@ -21,14 +23,32 @@ fn main() {
}
}
}
- let a = HerculesCPURef::from_slice(&a);
- let b = HerculesCPURef::from_slice(&b);
- let mut r = runner!(matmul);
- let c = r.run(I as u64, J as u64, K as u64, a.clone(), b.clone()).await;
- assert_eq!(c.as_slice::<i32>(), &*correct_c);
- let mut r = runner!(tiled_64_matmul);
- let tiled_c = r.run(I as u64, J as u64, K as u64, a.clone(), b.clone()).await;
- assert_eq!(tiled_c.as_slice::<i32>(), &*correct_c);
+ #[cfg(not(feature = "cuda"))]
+ {
+ let a = HerculesCPURef::from_slice(&a);
+ let b = HerculesCPURef::from_slice(&b);
+ let mut r = runner!(matmul);
+ let c = r.run(I as u64, J as u64, K as u64, a.clone(), b.clone()).await;
+ assert_eq!(c.as_slice::<i32>(), &*correct_c);
+ let mut r = runner!(tiled_64_matmul);
+ let tiled_c = r.run(I as u64, J as u64, K as u64, a.clone(), b.clone()).await;
+ assert_eq!(tiled_c.as_slice::<i32>(), &*correct_c);
+ }
+ #[cfg(feature = "cuda")]
+ {
+ let a = CUDABox::from_cpu_ref(HerculesCPURef::from_slice(&mut a));
+ let b = CUDABox::from_cpu_ref(HerculesCPURef::from_slice(&mut b));
+ let mut r = runner!(matmul);
+ let c = r.run(I as u64, J as u64, K as u64, a.get_ref(), b.get_ref()).await;
+ let mut c_cpu: Box<[i32]> = vec![0; correct_c.len()].into_boxed_slice();
+ c.to_cpu_ref(&mut c_cpu);
+ assert_eq!(&*c_cpu, &*correct_c);
+ let mut r = runner!(tiled_64_matmul);
+ let tiled_c = r.run(I as u64, J as u64, K as u64, a.get_ref(), b.get_ref()).await;
+ let mut tiled_c_cpu: Box<[i32]> = vec![0; correct_c.len()].into_boxed_slice();
+ tiled_c.to_cpu_ref(&mut tiled_c_cpu);
+ assert_eq!(&*tiled_c_cpu, &*correct_c);
+ }
});
}
@@ -36,4 +56,3 @@ fn main() {
fn matmul_test() {
main();
}
-
diff --git a/juno_samples/nested_ccp/build.rs b/juno_samples/nested_ccp/build.rs
index c5c7ca6a1b9ab5decf6a8cf0b8e8f13ff7122834..2352ddef0a432d42a09fef72042c516843799290 100644
--- a/juno_samples/nested_ccp/build.rs
+++ b/juno_samples/nested_ccp/build.rs
@@ -1,6 +1,15 @@
use juno_build::JunoCompiler;
fn main() {
+ #[cfg(feature = "cuda")]
+ JunoCompiler::new()
+ .file_in_src("nested_ccp.jn")
+ .unwrap()
+ .schedule_in_src("gpu.sch")
+ .unwrap()
+ .build()
+ .unwrap();
+ #[cfg(not(feature = "cuda"))]
JunoCompiler::new()
.file_in_src("nested_ccp.jn")
.unwrap()
diff --git a/juno_samples/nested_ccp/src/gpu.sch b/juno_samples/nested_ccp/src/gpu.sch
new file mode 100644
index 0000000000000000000000000000000000000000..021a05e3f361e4f5d1ae679c33f24282d431b2cc
--- /dev/null
+++ b/juno_samples/nested_ccp/src/gpu.sch
@@ -0,0 +1,18 @@
+gvn(*);
+phi-elim(*);
+dce(*);
+
+let out = auto-outline(*);
+gpu(out.ccp_example, out.median_array, out.no_underflow);
+
+ip-sroa(*);
+sroa(*);
+dce(*);
+float-collections(*);
+gvn(*);
+phi-elim(*);
+dce(*);
+
+infer-schedules(*);
+
+gcm(*);
diff --git a/juno_samples/nested_ccp/src/main.rs b/juno_samples/nested_ccp/src/main.rs
index 423b66fb8c84fd1af6b0267c7c63aa204ab1dc6c..412d56a4f4e66af5a6608822219bce16ac5554dc 100644
--- a/juno_samples/nested_ccp/src/main.rs
+++ b/juno_samples/nested_ccp/src/main.rs
@@ -1,6 +1,8 @@
#![feature(concat_idents)]
use hercules_rt::{runner, HerculesCPURef, HerculesCPURefMut};
+#[cfg(feature = "cuda")]
+use hercules_rt::CUDABox;
juno_build::juno!("nested_ccp");
@@ -8,19 +10,30 @@ fn main() {
async_std::task::block_on(async {
let a: Box<[f32]> = Box::new([17.0, 18.0, 19.0]);
let mut b: Box<[i32]> = Box::new([12, 16, 4, 18, 23, 56, 93, 22, 14]);
- let a = HerculesCPURef::from_slice(&a);
- let b = HerculesCPURefMut::from_slice(&mut b);
- let mut r = runner!(ccp_example);
- let output_example = r.run(a).await;
- let mut r = runner!(median_array);
- let output_median = r.run(9, b).await;
+ #[cfg(not(feature = "cuda"))]
+ {
+ let a = HerculesCPURef::from_slice(&a);
+ let b = HerculesCPURefMut::from_slice(&mut b);
+ let mut r = runner!(ccp_example);
+ let output_example = r.run(a).await;
+ let mut r = runner!(median_array);
+ let output_median = r.run(9, b).await;
+ assert_eq!(output_example, 1.0);
+ assert_eq!(output_median, 18);
+ }
+ #[cfg(feature = "cuda")]
+ {
+ let mut a = CUDABox::from_cpu_ref(HerculesCPURef::from_slice(&a));
+ let mut b = CUDABox::from_cpu_ref(HerculesCPURef::from_slice(&mut b));
+ let mut r = runner!(ccp_example);
+ let output_example = r.run(a.get_ref_mut()).await;
+ let mut r = runner!(median_array);
+ let output_median = r.run(9, b.get_ref_mut()).await;
+ assert_eq!(output_example, 1.0);
+ assert_eq!(output_median, 18);
+ }
let mut r = runner!(no_underflow);
let out_no_underflow = r.run().await;
- println!("{}", output_example);
- println!("{}", output_median);
- println!("{}", out_no_underflow);
- assert_eq!(output_example, 1.0);
- assert_eq!(output_median, 18);
assert_eq!(out_no_underflow, 7);
});
}
diff --git a/juno_samples/schedule_test/src/main.rs b/juno_samples/schedule_test/src/main.rs
index 2e63babf29e84bc74a7649306dc28af88225cf39..1505d4e5ff620a53d1095cdc4185a5a6d665e71e 100644
--- a/juno_samples/schedule_test/src/main.rs
+++ b/juno_samples/schedule_test/src/main.rs
@@ -3,6 +3,8 @@
use rand::random;
use hercules_rt::{runner, HerculesCPURef};
+#[cfg(feature = "cuda")]
+use hercules_rt::CUDABox;
juno_build::juno!("code");
@@ -26,12 +28,26 @@ fn main() {
}
}
- let a = HerculesCPURef::from_slice(&a);
- let b = HerculesCPURef::from_slice(&b);
- let c = HerculesCPURef::from_slice(&c);
- let mut r = runner!(test);
- let res = r.run(N as u64, M as u64, K as u64, a, b, c).await;
- assert_eq!(res.as_slice::<i32>(), &*correct_res);
+ #[cfg(not(feature = "cuda"))]
+ {
+ let a = HerculesCPURef::from_slice(&a);
+ let b = HerculesCPURef::from_slice(&b);
+ let c = HerculesCPURef::from_slice(&c);
+ let mut r = runner!(test);
+ let res = r.run(N as u64, M as u64, K as u64, a, b, c).await;
+ assert_eq!(res.as_slice::<i32>(), &*correct_res);
+ }
+ #[cfg(feature = "cuda")]
+ {
+ let a = CUDABox::from_cpu_ref(HerculesCPURef::from_slice(&a));
+ let b = CUDABox::from_cpu_ref(HerculesCPURef::from_slice(&b));
+ let c = CUDABox::from_cpu_ref(HerculesCPURef::from_slice(&c));
+ let mut r = runner!(test);
+ let res = r.run(N as u64, M as u64, K as u64, a.get_ref(), b.get_ref(), c.get_ref()).await;
+ let mut res_cpu: Box<[i32]> = vec![0; correct_res.len()].into_boxed_slice();
+ res.to_cpu_ref(&mut res_cpu);
+ assert_eq!(&*res_cpu, &*correct_res);
+ }
});
}
diff --git a/juno_samples/simple3/build.rs b/juno_samples/simple3/build.rs
index 94760025d53abe7e10914052e1a7783386b316b0..a0874af7ecb42b29eea3aaaa5628d1c0ddeb2090 100644
--- a/juno_samples/simple3/build.rs
+++ b/juno_samples/simple3/build.rs
@@ -1,6 +1,15 @@
use juno_build::JunoCompiler;
fn main() {
+ #[cfg(feature = "cuda")]
+ JunoCompiler::new()
+ .file_in_src("simple3.jn")
+ .unwrap()
+ .schedule_in_src("gpu.sch")
+ .unwrap()
+ .build()
+ .unwrap();
+ #[cfg(not(feature = "cuda"))]
JunoCompiler::new()
.file_in_src("simple3.jn")
.unwrap()
diff --git a/juno_samples/simple3/src/gpu.sch b/juno_samples/simple3/src/gpu.sch
new file mode 100644
index 0000000000000000000000000000000000000000..e97627d43c0e1dccf67f565c686712acb1f080f8
--- /dev/null
+++ b/juno_samples/simple3/src/gpu.sch
@@ -0,0 +1,18 @@
+gvn(*);
+phi-elim(*);
+dce(*);
+
+let out = auto-outline(*);
+gpu(out.simple3);
+
+ip-sroa(*);
+sroa(*);
+dce(*);
+float-collections(*);
+gvn(*);
+phi-elim(*);
+dce(*);
+
+infer-schedules(*);
+
+gcm(*);
diff --git a/juno_samples/simple3/src/main.rs b/juno_samples/simple3/src/main.rs
index 4f9fe6a708ec50fe65b1c31a2823580a117985ce..8eb78f7c93b0f195ffee1be120376dbe3f9a2a62 100644
--- a/juno_samples/simple3/src/main.rs
+++ b/juno_samples/simple3/src/main.rs
@@ -1,6 +1,8 @@
#![feature(concat_idents)]
use hercules_rt::{runner, HerculesCPURef};
+#[cfg(feature = "cuda")]
+use hercules_rt::CUDABox;
juno_build::juno!("simple3");
@@ -8,12 +10,22 @@ fn main() {
async_std::task::block_on(async {
let a: Box<[u32]> = Box::new([1, 2, 3, 4, 5, 6, 7, 8]);
let b: Box<[u32]> = Box::new([8, 7, 6, 5, 4, 3, 2, 1]);
- let a = HerculesCPURef::from_slice(&a);
- let b = HerculesCPURef::from_slice(&b);
- let mut r = runner!(simple3);
- let c = r.run(8, a, b).await;
- println!("{}", c);
- assert_eq!(c, 120);
+ #[cfg(not(feature = "cuda"))]
+ {
+ let a = HerculesCPURef::from_slice(&a);
+ let b = HerculesCPURef::from_slice(&b);
+ let mut r = runner!(simple3);
+ let c = r.run(8, a, b).await;
+ assert_eq!(c, 120);
+ }
+ #[cfg(feature = "cuda")]
+ {
+ let a = CUDABox::from_cpu_ref(HerculesCPURef::from_slice(&a));
+ let b = CUDABox::from_cpu_ref(HerculesCPURef::from_slice(&b));
+ let mut r = runner!(simple3);
+ let c = r.run(8, a.get_ref(), b.get_ref()).await;
+ assert_eq!(c, 120);
+ }
});
}