Thanks for using Compiler Explorer
Sponsors
Jakt
C++
Ada
Analysis
Android Java
Android Kotlin
Assembly
C
C3
Carbon
C++ (Circle)
CIRCT
Clean
CMake
CMakeScript
COBOL
C++ for OpenCL
MLIR
Cppx
Cppx-Blue
Cppx-Gold
Cpp2-cppfront
Crystal
C#
CUDA C++
D
Dart
Elixir
Erlang
Fortran
F#
Go
Haskell
HLSL
Hook
Hylo
ispc
Java
Julia
Kotlin
LLVM IR
LLVM MIR
Modula-2
Nim
Objective-C
Objective-C++
OCaml
OpenCL C
Pascal
Pony
Python
Racket
Ruby
Rust
Snowball
Scala
Solidity
Spice
Swift
LLVM TableGen
Toit
TypeScript Native
V
Vala
Visual Basic
WASM
Zig
Javascript
GIMPLE
rust source #1
Output
Compile to binary object
Link to binary
Execute the code
Intel asm syntax
Demangle identifiers
Verbose demangling
Filters
Unused labels
Library functions
Directives
Comments
Horizontal whitespace
Debug intrinsics
Compiler
mrustc (master)
rustc 1.0.0
rustc 1.1.0
rustc 1.10.0
rustc 1.11.0
rustc 1.12.0
rustc 1.13.0
rustc 1.14.0
rustc 1.15.1
rustc 1.16.0
rustc 1.17.0
rustc 1.18.0
rustc 1.19.0
rustc 1.2.0
rustc 1.20.0
rustc 1.21.0
rustc 1.22.0
rustc 1.23.0
rustc 1.24.0
rustc 1.25.0
rustc 1.26.0
rustc 1.27.0
rustc 1.27.1
rustc 1.28.0
rustc 1.29.0
rustc 1.3.0
rustc 1.30.0
rustc 1.31.0
rustc 1.32.0
rustc 1.33.0
rustc 1.34.0
rustc 1.35.0
rustc 1.36.0
rustc 1.37.0
rustc 1.38.0
rustc 1.39.0
rustc 1.4.0
rustc 1.40.0
rustc 1.41.0
rustc 1.42.0
rustc 1.43.0
rustc 1.44.0
rustc 1.45.0
rustc 1.45.2
rustc 1.46.0
rustc 1.47.0
rustc 1.48.0
rustc 1.49.0
rustc 1.5.0
rustc 1.50.0
rustc 1.51.0
rustc 1.52.0
rustc 1.53.0
rustc 1.54.0
rustc 1.55.0
rustc 1.56.0
rustc 1.57.0
rustc 1.58.0
rustc 1.59.0
rustc 1.6.0
rustc 1.60.0
rustc 1.61.0
rustc 1.62.0
rustc 1.63.0
rustc 1.64.0
rustc 1.65.0
rustc 1.66.0
rustc 1.67.0
rustc 1.68.0
rustc 1.69.0
rustc 1.7.0
rustc 1.70.0
rustc 1.71.0
rustc 1.72.0
rustc 1.73.0
rustc 1.74.0
rustc 1.75.0
rustc 1.76.0
rustc 1.77.0
rustc 1.78.0
rustc 1.79.0
rustc 1.8.0
rustc 1.80.0
rustc 1.81.0
rustc 1.9.0
rustc beta
rustc nightly
rustc-cg-gcc (master)
x86-64 GCCRS (GCC master)
x86-64 GCCRS (GCCRS master)
x86-64 GCCRS 14.1 (GCC)
x86-64 GCCRS 14.2 (GCC)
Options
Source code
use std::convert::TryInto; use std::arch::x86_64::*; #[target_feature(enable = "sse4.1")] pub unsafe fn transpose_8by8_sse4( in_matrix: &[u8], out: &mut [u8], in_stride: usize, out_stride: usize, ) { // Godbolt : https://godbolt.org/z/jan4WP1xr // Stack overflow: https://stackoverflow.com/a/42316675 assert!((8 * out_stride) <= out.len()); assert!((8 * in_stride) <= in_matrix.len()); let sv = _mm_set_epi8(15, 7, 14, 6, 13, 5, 12, 4, 11, 3, 10, 2, 9, 1, 8, 0); let mut pos = 0; // Load data from memory // Load 64 bites to ensure we only take 8 values let mn_0 = _mm_loadl_epi64(in_matrix.get_unchecked(pos..).as_ptr().cast()); pos += in_stride; let mn_1 = _mm_loadl_epi64(in_matrix.get_unchecked(pos..).as_ptr().cast()); pos += in_stride; let mv_0 = _mm_unpacklo_epi64(mn_0, mn_1); let mn_2 = _mm_loadl_epi64(in_matrix.get_unchecked(pos..).as_ptr().cast()); pos += in_stride; let mn_3 = _mm_loadl_epi64(in_matrix.get_unchecked(pos..).as_ptr().cast()); pos += in_stride; let mv_1 = _mm_unpacklo_epi64(mn_2, mn_3); let mn_4 = _mm_loadl_epi64(in_matrix.get_unchecked(pos..).as_ptr().cast()); pos += in_stride; let mn_5 = _mm_loadl_epi64(in_matrix.get_unchecked(pos..).as_ptr().cast()); pos += in_stride; let mv_2 = _mm_unpacklo_epi64(mn_4, mn_5); let mn_6 = _mm_loadl_epi64(in_matrix.get_unchecked(pos..).as_ptr().cast()); pos += in_stride; let mn_7 = _mm_loadl_epi64(in_matrix.get_unchecked(pos..).as_ptr().cast()); let mv_3 = _mm_unpacklo_epi64(mn_6, mn_7); let ov_0 = _mm_shuffle_epi8(mv_0, sv); let ov_1 = _mm_shuffle_epi8(mv_1, sv); let ov_2 = _mm_shuffle_epi8(mv_2, sv); let ov_3 = _mm_shuffle_epi8(mv_3, sv); let iv_0 = _mm_unpacklo_epi16(ov_0, ov_1); let iv_1 = _mm_unpackhi_epi16(ov_0, ov_1); let iv_2 = _mm_unpacklo_epi16(ov_2, ov_3); let iv_3 = _mm_unpackhi_epi16(ov_2, ov_3); let av_0 = _mm_unpacklo_epi32(iv_0, iv_2); let av_1 = _mm_unpackhi_epi32(iv_0, iv_2); let av_2 = _mm_unpacklo_epi32(iv_1, iv_3); let av_3 = _mm_unpackhi_epi32(iv_1, iv_3); // Now we have av1 having 0-16, av2 - 16-32 etc etc // So we want to extract and write only 8 bytes, as that is essentially a matrix // transpose of a 8 by 8 matrix writing to different strides. let sv_0 = _mm_unpackhi_epi64(av_0, _mm_setzero_si128()); let sv_1 = _mm_unpackhi_epi64(av_1, _mm_setzero_si128()); let sv_2 = _mm_unpackhi_epi64(av_2, _mm_setzero_si128()); let sv_3 = _mm_unpackhi_epi64(av_3, _mm_setzero_si128()); pos = 0; // Ensure writes are always in bounds // Needed to make the below writes unsafe _mm_storel_epi64(out.get_unchecked_mut(pos..).as_mut_ptr().cast(), av_0); pos += out_stride; _mm_storel_epi64(out.get_unchecked_mut(pos..).as_mut_ptr().cast(), sv_0); pos += out_stride; _mm_storel_epi64(out.get_unchecked_mut(pos..).as_mut_ptr().cast(), av_1); pos += out_stride; _mm_storel_epi64(out.get_unchecked_mut(pos..).as_mut_ptr().cast(), sv_1); pos += out_stride; _mm_storel_epi64(out.get_unchecked_mut(pos..).as_mut_ptr().cast(), av_2); pos += out_stride; _mm_storel_epi64(out.get_unchecked_mut(pos..).as_mut_ptr().cast(), sv_2); pos += out_stride; _mm_storel_epi64(out.get_unchecked_mut(pos..).as_mut_ptr().cast(), av_3); pos += out_stride; _mm_storel_epi64(out.get_unchecked_mut(pos..).as_mut_ptr().cast(), sv_3); } fn main() { let t = (0..192).collect::<Vec<u8>>(); let mut out = [0_u8; 64]; unsafe { transpose_8by8_sse4(&t, &mut out, 15, 8); } print!("["); for ch in out.chunks_exact(8) { println!("{:?}", ch); } }
Become a Patron
Sponsor on GitHub
Donate via PayPal
Source on GitHub
Mailing list
Installed libraries
Wiki
Report an issue
How it works
Contact the author
CE on Mastodon
About the author
Statistics
Changelog
Version tree