Benoit Jacob
Benoit Jacob
In SIMD instruction sets where integer arithmetic is really first-class, support for different bit width in different operands is not just an ad-hoc addition for a few instructions. Instead, generically...
Unlike the float case where the fused-vs-unfused issue creates complications (PR #79) in the integer case there is no downside to using single-instruction multiply-add. These are vital to getting above...
On Android/ARM64, Release config, size of `iree-run-module` (after stripping): Before: 794424 After: 591472
Take care of cases where one of the M, N or K dimension is 0, in early-return paths. This is motivated by loop pipelining optimizations that I want to do...
This set of benchmarks corresponds to the workloads that I'm tracking in ongoing work. This replaces the existing VMVX benchmarks. The comment there suggested that this was just about having...
FYI @wolfpld: Capstone has just done a breaking renaming: ARM64 --> AArch64 (https://github.com/capstone-engine/capstone/pull/2026) This PR is a good-enough local patch for anyone trying to build against latest Capstone. It uses...
Recently @mariecwhite has been adding `s8s4s32` code paths to the mmt4d ukernel, including optimized code paths for arm64 but not for x86-64. This Issue is about adding the x86-64 pieces....
There are two parts to this PR: * The `runtime/` changes bring a minimal weak-hook extension point around executable library calls. It's behind a C-internal flag: to enable it, configure...
I had a conversaion with @benvanik yesterday and for the first time I feel that I understand where we want to be and an incremental path to that. Some motivation...
Our build needs to know the path to Clang built-in headers to set up the compilation command line to build bitcode. That path is always of the form `clang/${LLVM_VERSION_MAJOR}/include`. In...