[PATCH] D118977: [NVPTX] Add more FMA intriniscs/builtins

This revision was landed with ongoing or failed builds. This revision was automatically updated to reflect the committed changes. Closed by commit rGbe672934ff88: [NVPTX] Add more FMA intriniscs/builtins (authored by jchlanda, committed by tra). Changed prior to commit: https://reviews.llvm.org

[PATCH] D118977: [NVPTX] Add more FMA intriniscs/builtins

tra added a comment. In D118977#3329146 , @jchlanda wrote: > @tra I've fixed the test failure (`math-intrins.ll`) the rest seems to be > unrelated timeouts, Thank you. > would you be able to merge those patches in, as I don't have the commit > access

[PATCH] D118977: [NVPTX] Add more FMA intriniscs/builtins

jchlanda added a comment. @tra I've fixed the test failure (`math-intrins.ll`) the rest seems to be unrelated timeouts, would you be able to merge those patches in, as I don't have the commit access please? The same goes for https://reviews.llvm.org/D117887 and https://reviews.llvm.org/D119157

[PATCH] D118977: [NVPTX] Add more FMA intriniscs/builtins

jchlanda updated this revision to Diff 407904. jchlanda added a comment. PTX/sm version tidy up. Repository: rG LLVM Github Monorepo CHANGES SINCE LAST ACTION https://reviews.llvm.org/D118977/new/ https://reviews.llvm.org/D118977 Files: clang/include/clang/Basic/BuiltinsNVPTX.def clan

[PATCH] D118977: [NVPTX] Add more FMA intriniscs/builtins

jchlanda added inline comments. Comment at: llvm/lib/Target/NVPTX/NVPTXIntrinsics.td:937 +class FMA_TUPLE Preds = [hasPTX70, hasSM80]> { + string Variant = V; tra wrote: > I think the default should be the most useful/common and the least > surprising value. >

[PATCH] D118977: [NVPTX] Add more FMA intriniscs/builtins

jchlanda updated this revision to Diff 407078. jchlanda added a comment. Tidy up FMA_TUPLE class. Repository: rG LLVM Github Monorepo CHANGES SINCE LAST ACTION https://reviews.llvm.org/D118977/new/ https://reviews.llvm.org/D118977 Files: clang/include/clang/Basic/BuiltinsNVPTX.def cla

[PATCH] D118977: [NVPTX] Add more FMA intriniscs/builtins

tra added inline comments. Comment at: llvm/lib/Target/NVPTX/NVPTXIntrinsics.td:937 +class FMA_TUPLE Preds = [hasPTX70, hasSM80]> { + string Variant = V; I think the default should be the most useful/common and the least surprising value. I'd argue that in thi

[PATCH] D118977: [NVPTX] Add more FMA intriniscs/builtins

jchlanda added a comment. In D118977#3302158 , @tra wrote: > In D118977#3299974 , @jchlanda > wrote: > >>> Target ISA Notes >>> Requires sm_53 or higher. > > I think we do need this constraint applied to the new

[PATCH] D118977: [NVPTX] Add more FMA intriniscs/builtins

jchlanda updated this revision to Diff 406851. jchlanda added a comment. Add sm/ptx version guard to f16{x2} builtins. Repository: rG LLVM Github Monorepo CHANGES SINCE LAST ACTION https://reviews.llvm.org/D118977/new/ https://reviews.llvm.org/D118977 Files: clang/include/clang/Basic/Bu

[PATCH] D118977: [NVPTX] Add more FMA intriniscs/builtins

tra added a comment. In D118977#3299974 , @jchlanda wrote: >> Target ISA Notes >> Requires sm_53 or higher. I think we do need this constraint applied to the new builtins, too. Right now nothing stops using them on a GPU where they do not exist and that

[PATCH] D118977: [NVPTX] Add more FMA intriniscs/builtins

jchlanda added a comment. In D118977#3297465 , @tra wrote: >> They all require PTX 7.0, SM_80. > > According to > https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#half-precision-floating-point-instructions-fma > only `fma.relu` and `bf1

[PATCH] D118977: [NVPTX] Add more FMA intriniscs/builtins

jchlanda updated this revision to Diff 406322. jchlanda added a comment. Set correct SM and PTX version. Repository: rG LLVM Github Monorepo CHANGES SINCE LAST ACTION https://reviews.llvm.org/D118977/new/ https://reviews.llvm.org/D118977 Files: clang/include/clang/Basic/BuiltinsNVPTX.de

[PATCH] D118977: [NVPTX] Add more FMA intriniscs/builtins

tra added a comment. > They all require PTX 7.0, SM_80. According to https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#half-precision-floating-point-instructions-fma only `fma.relu` and `bf16*` variants require ptx70/sm80: PTX ISA Notes Introduced in PTX ISA version 4.2.

[PATCH] D118977: [NVPTX] Add more FMA intriniscs/builtins

jchlanda created this revision. jchlanda added reviewers: tra, csigg, bkramer. Herald added subscribers: asavonic, hiraditya, jholewinski. jchlanda requested review of this revision. Herald added subscribers: llvm-commits, cfe-commits, jdoerfert. Herald added projects: clang, LLVM. This patch adds