mikros/rust
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
master
rust/tests/codegen/array-codegen.rs

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

63 lines
2.3 KiB
Rust
Raw Permalink Normal View History

Use `load`-`store` instead of `memcpy` for short integer arrays
2023-05-26 05:32:22 -05:00
//@ compile-flags: -O -C no-prepopulate-passes
#![crate_type = "lib"]
// CHECK-LABEL: @array_load
#[no_mangle]
pub fn array_load(a: &[u8; 4]) -> [u8; 4] {
Remove my `scalar_copy_backend_type` optimization attempt I added this back in 111999, but I no longer think it's a good idea - It had to get scaled back to only power-of-two things to not break a bunch of targets - LLVM seems to be getting better at memcpy removal anyway - Introducing vector instructions has seemed to sometimes (115515) make autovectorization worse So this removes it from the codegen crates entirely, and instead just tries to use <https://doc.rust-lang.org/nightly/nightly-rustc/rustc_codegen_ssa/traits/builder/trait.BuilderMethods.html#method.typed_place_copy> instead of direct `memcpy` so things will still use load/store for immediates.
2024-03-29 02:00:24 -05:00
// CHECK-NOT: alloca
// CHECK: %[[ALLOCA:.+]] = alloca [4 x i8], align 1
// CHECK-NOT: alloca
// CHECK: call void @llvm.memcpy.{{.+}}(ptr align 1 %[[ALLOCA]], ptr align 1 %a, {{.+}} 4, i1 false)
// CHECK: %[[TEMP:.+]] = load i32, ptr %[[ALLOCA]], align 1
// CHECK: ret i32 %[[TEMP]]
Use `load`-`store` instead of `memcpy` for short integer arrays
2023-05-26 05:32:22 -05:00
*a
}
// CHECK-LABEL: @array_store
#[no_mangle]
pub fn array_store(a: [u8; 4], p: &mut [u8; 4]) {
Remove my `scalar_copy_backend_type` optimization attempt I added this back in 111999, but I no longer think it's a good idea - It had to get scaled back to only power-of-two things to not break a bunch of targets - LLVM seems to be getting better at memcpy removal anyway - Introducing vector instructions has seemed to sometimes (115515) make autovectorization worse So this removes it from the codegen crates entirely, and instead just tries to use <https://doc.rust-lang.org/nightly/nightly-rustc/rustc_codegen_ssa/traits/builder/trait.BuilderMethods.html#method.typed_place_copy> instead of direct `memcpy` so things will still use load/store for immediates.
2024-03-29 02:00:24 -05:00
// CHECK-NOT: alloca
use [N x i8] for alloca types
2024-02-23 23:48:20 -06:00
// CHECK: %[[TEMP:.+]] = alloca [4 x i8], [[TEMPALIGN:align [0-9]+]]
Remove my `scalar_copy_backend_type` optimization attempt I added this back in 111999, but I no longer think it's a good idea - It had to get scaled back to only power-of-two things to not break a bunch of targets - LLVM seems to be getting better at memcpy removal anyway - Introducing vector instructions has seemed to sometimes (115515) make autovectorization worse So this removes it from the codegen crates entirely, and instead just tries to use <https://doc.rust-lang.org/nightly/nightly-rustc/rustc_codegen_ssa/traits/builder/trait.BuilderMethods.html#method.typed_place_copy> instead of direct `memcpy` so things will still use load/store for immediates.
2024-03-29 02:00:24 -05:00
// CHECK-NOT: alloca
Use `load`-`store` instead of `memcpy` for short integer arrays
2023-05-26 05:32:22 -05:00
// CHECK: %a = alloca [4 x i8]
Remove my `scalar_copy_backend_type` optimization attempt I added this back in 111999, but I no longer think it's a good idea - It had to get scaled back to only power-of-two things to not break a bunch of targets - LLVM seems to be getting better at memcpy removal anyway - Introducing vector instructions has seemed to sometimes (115515) make autovectorization worse So this removes it from the codegen crates entirely, and instead just tries to use <https://doc.rust-lang.org/nightly/nightly-rustc/rustc_codegen_ssa/traits/builder/trait.BuilderMethods.html#method.typed_place_copy> instead of direct `memcpy` so things will still use load/store for immediates.
2024-03-29 02:00:24 -05:00
// CHECK-NOT: alloca
// store i32 %0, ptr %[[TEMP]]
// CHECK: call void @llvm.memcpy.{{.+}}(ptr align 1 %a, ptr [[TEMPALIGN]] %[[TEMP]], {{.+}} 4, i1 false)
// CHECK: call void @llvm.memcpy.{{.+}}(ptr align 1 %p, ptr align 1 %a, {{.+}} 4, i1 false)
Use `load`-`store` instead of `memcpy` for short integer arrays
2023-05-26 05:32:22 -05:00
*p = a;
}
// CHECK-LABEL: @array_copy
#[no_mangle]
pub fn array_copy(a: &[u8; 4], p: &mut [u8; 4]) {
Remove my `scalar_copy_backend_type` optimization attempt I added this back in 111999, but I no longer think it's a good idea - It had to get scaled back to only power-of-two things to not break a bunch of targets - LLVM seems to be getting better at memcpy removal anyway - Introducing vector instructions has seemed to sometimes (115515) make autovectorization worse So this removes it from the codegen crates entirely, and instead just tries to use <https://doc.rust-lang.org/nightly/nightly-rustc/rustc_codegen_ssa/traits/builder/trait.BuilderMethods.html#method.typed_place_copy> instead of direct `memcpy` so things will still use load/store for immediates.
2024-03-29 02:00:24 -05:00
// CHECK-NOT: alloca
Use `load`-`store` instead of `memcpy` for short integer arrays
2023-05-26 05:32:22 -05:00
// CHECK: %[[LOCAL:.+]] = alloca [4 x i8], align 1
Remove my `scalar_copy_backend_type` optimization attempt I added this back in 111999, but I no longer think it's a good idea - It had to get scaled back to only power-of-two things to not break a bunch of targets - LLVM seems to be getting better at memcpy removal anyway - Introducing vector instructions has seemed to sometimes (115515) make autovectorization worse So this removes it from the codegen crates entirely, and instead just tries to use <https://doc.rust-lang.org/nightly/nightly-rustc/rustc_codegen_ssa/traits/builder/trait.BuilderMethods.html#method.typed_place_copy> instead of direct `memcpy` so things will still use load/store for immediates.
2024-03-29 02:00:24 -05:00
// CHECK-NOT: alloca
// CHECK: call void @llvm.memcpy.{{.+}}(ptr align 1 %[[LOCAL]], ptr align 1 %a, {{.+}} 4, i1 false)
// CHECK: call void @llvm.memcpy.{{.+}}(ptr align 1 %p, ptr align 1 %[[LOCAL]], {{.+}} 4, i1 false)
Use `load`-`store` instead of `memcpy` for short integer arrays
2023-05-26 05:32:22 -05:00
*p = *a;
}
Copy 1-element arrays as scalars, not vectors For `[T; 1]` it's silly to copy as `<1 x T>` when we can just copy as `T`.
2023-10-06 03:17:09 -05:00
// CHECK-LABEL: @array_copy_1_element
#[no_mangle]
pub fn array_copy_1_element(a: &[u8; 1], p: &mut [u8; 1]) {
Remove my `scalar_copy_backend_type` optimization attempt I added this back in 111999, but I no longer think it's a good idea - It had to get scaled back to only power-of-two things to not break a bunch of targets - LLVM seems to be getting better at memcpy removal anyway - Introducing vector instructions has seemed to sometimes (115515) make autovectorization worse So this removes it from the codegen crates entirely, and instead just tries to use <https://doc.rust-lang.org/nightly/nightly-rustc/rustc_codegen_ssa/traits/builder/trait.BuilderMethods.html#method.typed_place_copy> instead of direct `memcpy` so things will still use load/store for immediates.
2024-03-29 02:00:24 -05:00
// CHECK-NOT: alloca
Copy 1-element arrays as scalars, not vectors For `[T; 1]` it's silly to copy as `<1 x T>` when we can just copy as `T`.
2023-10-06 03:17:09 -05:00
// CHECK: %[[LOCAL:.+]] = alloca [1 x i8], align 1
Remove my `scalar_copy_backend_type` optimization attempt I added this back in 111999, but I no longer think it's a good idea - It had to get scaled back to only power-of-two things to not break a bunch of targets - LLVM seems to be getting better at memcpy removal anyway - Introducing vector instructions has seemed to sometimes (115515) make autovectorization worse So this removes it from the codegen crates entirely, and instead just tries to use <https://doc.rust-lang.org/nightly/nightly-rustc/rustc_codegen_ssa/traits/builder/trait.BuilderMethods.html#method.typed_place_copy> instead of direct `memcpy` so things will still use load/store for immediates.
2024-03-29 02:00:24 -05:00
// CHECK-NOT: alloca
// CHECK: call void @llvm.memcpy.{{.+}}(ptr align 1 %[[LOCAL]], ptr align 1 %a, {{.+}} 1, i1 false)
// CHECK: call void @llvm.memcpy.{{.+}}(ptr align 1 %p, ptr align 1 %[[LOCAL]], {{.+}} 1, i1 false)
Copy 1-element arrays as scalars, not vectors For `[T; 1]` it's silly to copy as `<1 x T>` when we can just copy as `T`.
2023-10-06 03:17:09 -05:00
*p = *a;
}
// CHECK-LABEL: @array_copy_2_elements
#[no_mangle]
pub fn array_copy_2_elements(a: &[u8; 2], p: &mut [u8; 2]) {
Remove my `scalar_copy_backend_type` optimization attempt I added this back in 111999, but I no longer think it's a good idea - It had to get scaled back to only power-of-two things to not break a bunch of targets - LLVM seems to be getting better at memcpy removal anyway - Introducing vector instructions has seemed to sometimes (115515) make autovectorization worse So this removes it from the codegen crates entirely, and instead just tries to use <https://doc.rust-lang.org/nightly/nightly-rustc/rustc_codegen_ssa/traits/builder/trait.BuilderMethods.html#method.typed_place_copy> instead of direct `memcpy` so things will still use load/store for immediates.
2024-03-29 02:00:24 -05:00
// CHECK-NOT: alloca
Copy 1-element arrays as scalars, not vectors For `[T; 1]` it's silly to copy as `<1 x T>` when we can just copy as `T`.
2023-10-06 03:17:09 -05:00
// CHECK: %[[LOCAL:.+]] = alloca [2 x i8], align 1
Remove my `scalar_copy_backend_type` optimization attempt I added this back in 111999, but I no longer think it's a good idea - It had to get scaled back to only power-of-two things to not break a bunch of targets - LLVM seems to be getting better at memcpy removal anyway - Introducing vector instructions has seemed to sometimes (115515) make autovectorization worse So this removes it from the codegen crates entirely, and instead just tries to use <https://doc.rust-lang.org/nightly/nightly-rustc/rustc_codegen_ssa/traits/builder/trait.BuilderMethods.html#method.typed_place_copy> instead of direct `memcpy` so things will still use load/store for immediates.
2024-03-29 02:00:24 -05:00
// CHECK-NOT: alloca
// CHECK: call void @llvm.memcpy.{{.+}}(ptr align 1 %[[LOCAL]], ptr align 1 %a, {{.+}} 2, i1 false)
// CHECK: call void @llvm.memcpy.{{.+}}(ptr align 1 %p, ptr align 1 %[[LOCAL]], {{.+}} 2, i1 false)
Copy 1-element arrays as scalars, not vectors For `[T; 1]` it's silly to copy as `<1 x T>` when we can just copy as `T`.
2023-10-06 03:17:09 -05:00
*p = *a;
}
Reference in New Issue Copy Permalink