rust/compiler/rustc_codegen_llvm/src/back, branch tryhttps://github.com/rust-lang/rust
http://git.dreamy.place/mirrors/rust/atom?h=try2025-07-21T16:54:24+00:00Rollup merge of #142097 - ZuseZ4:offload-host1, r=oli-obk2025-07-21T16:54:24+00:00许杰友 Jieyou Xu (Joe)39484203+jieyouxu@users.noreply.github.com2025-07-21T16:54:24+00:00urn:sha1:5e3eb2512591df0cef52404f0ea4202f58935a54
gpu offload host code generation
r? ghost
This will generate most of the host side code to use llvm's offload feature.
The first PR will only handle automatic mem-transfers to and from the device.
So if a user calls a kernel, we will copy inputs back and forth, but we won't do the actual kernel launch.
Before merging, we will use LLVM's Info infrastructure to verify that the memcopies match what openmp offloa generates in C++. `LIBOMPTARGET_INFO=-1 ./my_rust_binary` should print that a memcpy to and later from the device is happening.
A follow-up PR will generate the actual device-side kernel which will then do computations on the GPU.
A third PR will implement manual host2device and device2host functionality, but the goal is to minimize cases where a user has to overwrite our default handling due to performance issues.
I'm trying to get a full MVP out first, so this just recognizes GPU functions based on magic names. The final frontend will obviously move this over to use proper macros, like I'm already doing it for the autodiff work.
This work will also be compatible with std::autodiff, so one can differentiate GPU kernels.
Tracking:
- https://github.com/rust-lang/rust/issues/131513
gpu host code generation2025-07-18T23:30:42+00:00Manuel Drehwaldgit@manuel.drehwald.info2025-07-02T23:36:30+00:00urn:sha1:4a1a5a42952d05533fd4309ad0f3fe290abbf57cadd -Zoffload=Enable flag behind -Zunstable-options, to enable gpu (host) code generation2025-07-18T23:24:00+00:00Manuel Drehwaldgit@manuel.drehwald.info2025-06-18T22:29:43+00:00urn:sha1:634016478ec95c6ff933d32789e663ace78e8f82Pass wasm exception model to TargetOptions2025-07-18T07:35:50+00:00Nikita Popovnpopov@redhat.com2025-07-11T08:11:03+00:00urn:sha1:12b19be741ea07934d7478bd8e450dca8f85afe5
This is no longer implied by -wasm-enable-eh.
Rollup merge of #143388 - bjorn3:lto_refactors, r=compiler-errors2025-07-17T01:58:28+00:00León Orell Valerian Liehrme@fmease.dev2025-07-17T01:58:28+00:00urn:sha1:be5f8f299dce5c04e2a644546e780d8a07b0b14f
Various refactors to the LTO handling code
In particular reducing the sharing of code paths between fat and thin-LTO and making the fat LTO implementation more self-contained. This also moves some autodiff handling out of cg_ssa into cg_llvm given that Enzyme only works with LLVM anyway and an implementation for another backend may do things entirely differently. This will also make it a bit easier to split LTO handling out of the coordinator thread main loop into a separate loop, which should reduce the complexity of the coordinator thread.
Avoid a bunch of unnecessary `unsafe` blocks in cg_llvm2025-07-14T08:27:08+00:00Oli Scherergithub333195615777966@oli-obk.de2025-07-09T09:12:42+00:00urn:sha1:d3d51b4fdbd6854da015f501e6566ca17cb023e5Rollup merge of #143633 - dillona:noinline-assert, r=fee1-dead2025-07-11T17:45:23+00:00Matthias Krüger476013+matthiaskrgr@users.noreply.github.com2025-07-11T17:45:23+00:00urn:sha1:7f3204f34d20e61a7581d8c9b52d00bf832a74bf
fix: correct assertion to check for 'noinline' attribute presence before removal
Make some "safe" llvm ops actually sound2025-07-10T07:27:41+00:00Oli Scherergithub333195615777966@oli-obk.de2025-07-10T07:27:41+00:00urn:sha1:84eeca2e2fc40bb8d6641846f18af9d8fc6a9681fix: correct assertion to check for 'noinline' attribute presence before removal2025-07-08T11:24:50+00:00Dillon Amburgeydillona@dillona.com2025-07-08T11:18:18+00:00urn:sha1:b2299e20b2c2a25254fb554ccc59a8b60fdc25e0Move dcx creation into WriteBackendMethods::codegen2025-07-03T14:43:09+00:00bjorn317426603+bjorn3@users.noreply.github.com2025-07-03T14:43:09+00:00urn:sha1:7fd78df3463c761a693ea79265a6a1faaf2ed51b