[clang] [llvm] [OpenMP] Add parser/semantic support for dyn_groupprivate clause (PR #152651)

https://github.com/kevinsala edited https://github.com/llvm/llvm-project/pull/152651 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [OpenMP] Add parser/semantic support for dyn_groupprivate clause (PR #152651)

https://github.com/kevinsala edited https://github.com/llvm/llvm-project/pull/152651 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [OpenMP][Offload] Add support for dyn_groupprivate clause (PR #152651)

https://github.com/kevinsala updated https://github.com/llvm/llvm-project/pull/152651 >From 099c502bdf02ed9bc34bbfc70a6e786746ecee90 Mon Sep 17 00:00:00 2001 From: Kevin Sala Date: Fri, 8 Aug 2025 10:43:52 -0700 Subject: [PATCH] [OpenMP] Add parser/semantic support for dyn_groupprivate clause

[clang] [flang] [llvm] [OpenMP] Add parser/semantic support for dyn_groupprivate clause (PR #152651)

https://github.com/kevinsala updated https://github.com/llvm/llvm-project/pull/152651 >From 099c502bdf02ed9bc34bbfc70a6e786746ecee90 Mon Sep 17 00:00:00 2001 From: Kevin Sala Date: Fri, 8 Aug 2025 10:43:52 -0700 Subject: [PATCH 1/2] [OpenMP] Add parser/semantic support for dyn_groupprivate cla

[clang] [flang] [llvm] [OpenMP] Add parser/semantic support for dyn_groupprivate clause (PR #152651)

@@ -1104,6 +1107,7 @@ def OMP_Target : Directive<[Spelling<"target">]> { let allowedOnceClauses = [ VersionedClause, VersionedClause, +VersionedClause, kevinsala wrote: The restriction in the spec is correct. Currently, only `cgroup` is allowed i

[clang] [flang] [llvm] [OpenMP] Add parser/semantic support for dyn_groupprivate clause (PR #152651)

kevinsala wrote: > Has `dyn_groupprivate` already been approved to be in 6.1 or is this just a > PoC implementation? @shiltian It has been accepted for OpenMP 6.1. Currently I'm missing to restrict this clause to the 6.1 version. https://github.com/llvm/llvm-project/pull/152651 __

[clang] [llvm] [clang][CUDA] Avoid accounting for tail padding in LLVM offloading (PR #156229)

kevinsala wrote: I believe I can use the same size (without the trailing padding) for the amdgpu plugin. However, I have to remove the whole check below: https://github.com/llvm/llvm-project/blob/8edb5b4fb3eca44d16325a119257632e1ba003be/offload/plugins-nextgen/amdgpu/src/rtl.cpp#L3660 The chec

[clang] [llvm] [clang][CUDA] Avoid accounting for tail padding in LLVM offloading (PR #156229)

https://github.com/kevinsala created https://github.com/llvm/llvm-project/pull/156229 It seems that `cuLaunchKernel` expects the arguments size (`CU_LAUNCH_PARAM_BUFFER_SIZE`) without accounting for tail padding. For example, for a kernel with arguments `int *A, short B`, the function requires

[clang] [llvm] [clang][CUDA] Avoid accounting for tail padding in LLVM offloading (PR #156229)

https://github.com/kevinsala edited https://github.com/llvm/llvm-project/pull/156229 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [llvm] [clang][CUDA] Avoid accounting for tail padding in LLVM offloading (PR #156229)

@@ -351,12 +353,24 @@ Address CGNVCUDARuntime::prepareKernelArgsLLVMOffload(CodeGenFunction &CGF, "kernel_launch_params"); auto KernelArgsSize = CGM.getDataLayout().getTypeAllocSize(KernelArgsTy); + + // Avoid accounting the tail padding for CUDA. + auto KernelArgsS

[clang] [llvm] [clang][CUDA] Avoid accounting for tail padding in LLVM offloading (PR #156229)

kevinsala wrote: My understanding is that the CUDA Driver API requires the arguments in the buffer to be placed with the proper alignment (i.e., padding between fields). However, the trailing padding after the last element should not be accounted. Otherwise, if it's accounted, the `cuLaunchKer

[clang] [flang] [llvm] [OpenMP] Add parser/semantic support for dyn_groupprivate clause (PR #152651)

https://github.com/kevinsala updated https://github.com/llvm/llvm-project/pull/152651 >From 099c502bdf02ed9bc34bbfc70a6e786746ecee90 Mon Sep 17 00:00:00 2001 From: Kevin Sala Date: Fri, 8 Aug 2025 10:43:52 -0700 Subject: [PATCH 1/6] [OpenMP] Add parser/semantic support for dyn_groupprivate cla

[clang] [docs][OpenMP] Add docs section for OpenMP 6.1 implementation status (PR #155651)

https://github.com/kevinsala closed https://github.com/llvm/llvm-project/pull/155651 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [docs][OpenMP] Claim C/C++ implementation of dyn_groupprivate clause (PR #155653)

https://github.com/kevinsala edited https://github.com/llvm/llvm-project/pull/155653 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [docs][OpenMP] Claim C/C++ implementation of dyn_groupprivate clause (PR #155653)

https://github.com/kevinsala updated https://github.com/llvm/llvm-project/pull/155653 >From 3cf3c63bdfd3b518ad824f51b2501e0599415881 Mon Sep 17 00:00:00 2001 From: Kevin Sala Date: Wed, 27 Aug 2025 09:55:31 -0700 Subject: [PATCH] [docs][OpenMP] Claim C/C++ implementation of dyn_groupprivate cl

[clang] [docs][OpenMP] Claim C/C++ implementation of dyn_groupprivate clause (PR #155653)

https://github.com/kevinsala closed https://github.com/llvm/llvm-project/pull/155653 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

[clang] [flang] [llvm] [OpenMP] Add parser/semantic support for dyn_groupprivate clause (PR #152651)

@@ -1104,6 +1107,7 @@ def OMP_Target : Directive<[Spelling<"target">]> { let allowedOnceClauses = [ VersionedClause, VersionedClause, +VersionedClause, kevinsala wrote: I'll rebase and include your changes in #154549 once merged. https://github.

[clang] [docs][OpenMP] Add docs section for OpenMP 6.1 implementation status (PR #155651)

https://github.com/kevinsala updated https://github.com/llvm/llvm-project/pull/155651 >From d816064c37982addc287bc73c272b4f3c19a9882 Mon Sep 17 00:00:00 2001 From: Kevin Sala Date: Wed, 27 Aug 2025 09:30:14 -0700 Subject: [PATCH] [docs][OpenMP] Add docs section for OpenMP 6.1 implementation st

[clang] [llvm] [clang][CUDA] Avoid accounting for tail padding in LLVM offloading (PR #156229)

kevinsala wrote: @jhuber6 @Artem-B does the new approach look better? We now pass the same size for both nvidia and amdgpu kernels. https://github.com/llvm/llvm-project/pull/156229 ___ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.

[clang] [llvm] [clang][CUDA] Avoid accounting for tail padding in LLVM offloading (PR #156229)

@@ -3655,11 +3655,6 @@ Error AMDGPUKernelTy::launchImpl(GenericDeviceTy &GenericDevice, KernelArgsTy &KernelArgs, KernelLaunchParamsTy LaunchParams, AsyncInfoWrapperTy &AsyncInfo

[clang] [llvm] [clang][CUDA] Avoid accounting for tail padding in LLVM offloading (PR #156229)

@@ -3655,11 +3655,6 @@ Error AMDGPUKernelTy::launchImpl(GenericDeviceTy &GenericDevice, KernelArgsTy &KernelArgs, KernelLaunchParamsTy LaunchParams, AsyncInfoWrapperTy &AsyncInfo

[clang] [llvm] [Offload][PGO] Fix PGO on NVPTX targets (PR #143568)

@@ -947,11 +954,18 @@ bool InstrLowerer::lower() { if (!ContainsProfiling && !CoverageNamesVar) return MadeChange; + // Cached info for generating delayed offset calculations + // This is only relevant on NVPTX targets + SmallVector Kernels; + SmallVector ValueSites;

[clang] [llvm] [clang][CUDA] Avoid accounting for tail padding in LLVM offloading (PR #156229)

@@ -3655,11 +3655,6 @@ Error AMDGPUKernelTy::launchImpl(GenericDeviceTy &GenericDevice, KernelArgsTy &KernelArgs, KernelLaunchParamsTy LaunchParams, AsyncInfoWrapperTy &AsyncInfo