diff options
| author | CarolineConcatto <51754594+CarolineConcatto@users.noreply.github.com> | 2023-11-21 14:02:18 +0000 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2023-11-21 14:02:18 +0000 |
| commit | f79676a17eae4c63318561ba35613d97053fa12c (patch) | |
| tree | ca9668f8a4edd9d5b8893042a541d5b59cd72182 /llvm/lib/Analysis/BasicAliasAnalysis.cpp | |
| parent | ad86d3e94f21fb13e504b0b1044ef7c10bc86a98 (diff) | |
| download | llvm-f79676a17eae4c63318561ba35613d97053fa12c.zip llvm-f79676a17eae4c63318561ba35613d97053fa12c.tar.gz llvm-f79676a17eae4c63318561ba35613d97053fa12c.tar.bz2 | |
[SVE2.1][Clang][LLVM]Add BFloat16 builtin in Clang and LLVM intrinisc (#70362)
This patch implements the builtins in Clang
and the LLVM-IR intrinsic for the following:
For BFADD , BFSUB, BFMAX, BFMIN, BFMAXNM and BFMINNM, for instance:
svbfloat16_t svadd[_bf16]_m (svbool_t pg, svbfloat16_t zdn, svbfloat16_t
zm);
svbfloat16_t svadd[_bf16]_x (svbool_t pg, svbfloat16_t zdn, svbfloat16_t
zm);
svbfloat16_t svadd[_bf16]_z (svbool_t pg, svbfloat16_t zdn, svbfloat16_t
zm);
svbfloat16_t svadd[_n_bf16]_m (svbool_t pg, svbfloat16_t zdn, bfloat16_t
zm);
svbfloat16_t svadd[_n_bf16]_x (svbool_t pg, svbfloat16_t zdn, bfloat16_t
zm);
svbfloat16_t svadd[_n_bf16]_z (svbool_t pg, svbfloat16_t zdn, bfloat16_t
zm);
the add, could be replaced by sub, max, min, maxnm and minnm.
For BFMUL:
svbfloat16_t svmul[_bf16]_m(svbool_t pg, svbfloat16_t zdn, svbfloat16_t
zm);
svbfloat16_t svmul[_bf16]_x(svbool_t pg, svbfloat16_t zdn, svbfloat16_t
zm);
svbfloat16_t svmul[_bf16]_z(svbool_t pg, svbfloat16_t zdn, svbfloat16_t
zm);
svbfloat16_t svmul[_n_bf16]_m(svbool_t pg, svbfloat16_t zdn, bfloat16_t
zm);
svbfloat16_t svmul[_n_bf16]_x(svbool_t pg, svbfloat16_t zdn, bfloat16_t
zm);
svbfloat16_t svmul[_n_bf16]_z(svbool_t pg, svbfloat16_t zdn, bfloat16_t
zm);
svbfloat16_t svmul_lane[_bf16](svbfloat16_t zn, svbfloat16_t zm,
uint64_t imm_idx);
For BFCLAMP:
svbfloat16_t svclamp[_bf16](svbfloat16_t op, svbfloat16_t min,
svbfloat16_t max);
For BFMLA and BFMLS
svbfloat16_t svmla[_bf16]_m(svbool_t pg, svbfloat16_t zda, svbfloat16_t
zn,
svbfloat16_t zm);
svbfloat16_t svmla[_bf16]_z(svbool_t pg, svbfloat16_t zda, svbfloat16_t
zn,
svbfloat16_t zm);
svbfloat16_t svmla[_bf16]_x(svbool_t pg, svbfloat16_t zda, svbfloat16_t
zn,
svbfloat16_t zm);
svbfloat16_t svmla[_n_bf16]_m(svbool_t pg, svbfloat16_t zda,
svbfloat16_t zn,
bfloat16_t zm);
svbfloat16_t svmla[_n_bf16]_z(svbool_t pg, svbfloat16_t zda,
svbfloat16_t zn,
bfloat16_t zm);
svbfloat16_t svmla[_n_bf16]_x(svbool_t pg, svbfloat16_t zda,
svbfloat16_t zn,
bfloat16_t zm);
svbfloat16_t svmla_lane[_bf16](svbfloat16_t zda, svbfloat16_t zn,
svbfloat16_t zm, uint64_t
According to the PR#257[1]
[1]ARM-software/acle#257
Co-authored-by: Matthew Devereau <matthew.devereau@arm.com>
Diffstat (limited to 'llvm/lib/Analysis/BasicAliasAnalysis.cpp')
0 files changed, 0 insertions, 0 deletions