Vector-reduction arithmetic accepts vectors as inputs and produces scalars as outputs.
This class of vector operation forms the basis of many scientific computations.
In vector-reduction arithmetic, the evaluation off is independent of the order of the input elements of V.
Details
Details
- Reviewers
craig.topper igorb aymanmus - Commits
- rGfacb37cabf6b: [X86][AVX512][Clang][Intrinsics][reduce] Adding missing reduce (Operators: +,*…
rG98cb041891b3: [X86][AVX512][Clang][Intrinsics][reduce] Adding missing reduce (Operators: +,*…
rC285054: [X86][AVX512][Clang][Intrinsics][reduce] Adding missing reduce (Operators: +,*…
rC284963: [X86][AVX512][Clang][Intrinsics][reduce] Adding missing reduce (Operators: +,*…
rL285054: [X86][AVX512][Clang][Intrinsics][reduce] Adding missing reduce (Operators: +,*…
rL284963: [X86][AVX512][Clang][Intrinsics][reduce] Adding missing reduce (Operators: +,*…
Diff Detail
Diff Detail
- Repository
- rL LLVM
Event Timeline
lib/Headers/avx512fintrin.h | ||
---|---|---|
9677 ↗ | (On Diff #74408) | Please refactor the macro and change to _mm512_reduce_operator_64bit |
9733 ↗ | (On Diff #74408) | you can use _mm512_reduce_operator_double macro here. |
9783 ↗ | (On Diff #74408) | please change to _mm512_reduce_operator_32bit nad use it to implement masked version |
lib/Headers/avx512fintrin.h | ||
---|---|---|
9896 ↗ | (On Diff #74563) | Intel docs show the argument order being reverse of this. mask16 then m512. |
lib/Headers/avx512fintrin.h | ||
---|---|---|
9895 ↗ | (On Diff #74563) | Shouldn't this return float not int? |
Comment Actions
ping
lib/Headers/avx512fintrin.h | ||
---|---|---|
9895 ↗ | (On Diff #74563) | You are right, this should return float. |
lib/Headers/avx512fintrin.h | ||
---|---|---|
9705 ↗ | (On Diff #75302) | Should this be "long long"? I think's that what we use for 64-bit everywhere else in the intrinsic headers. Sorry I didn't notice the first time. |
9805 ↗ | (On Diff #75302) | I wonder if "0, 1, -1, -1" would be better here. |
9810 ↗ | (On Diff #75302) | I wonder if "2, 3, -1, -1" would be better here. |
9814 ↗ | (On Diff #75302) | I wonder if "0, -1, -1, -1" would be better here. |
9819 ↗ | (On Diff #75302) | I wonder if "1, -1, -1, -1" would be better here. |
lib/Headers/avx512fintrin.h | ||
---|---|---|
9805 ↗ | (On Diff #75302) | its better!!! |