diff options
| author | Michalis Spyrou <michalis.spyrou@arm.com> | 2019-01-14 17:27:39 +0000 |
|---|---|---|
| committer | Michalis Spyrou <michalis.spyrou@arm.com> | 2019-01-15 09:58:08 +0000 |
| commit | 254a48a303ab0a8497849d7d8b4d2ad3ab88461a (patch) | |
| tree | b100da4e5893707121efaffb919104783c0708a1 /src/core/NEON/kernels/NEReductionOperationKernel.cpp | |
| parent | 5a5945387e70f62e6e1e95a177fae261d7570443 (diff) | |
| download | ComputeLibrary-254a48a303ab0a8497849d7d8b4d2ad3ab88461a.tar.gz | |
COMPMID-1885 (Nightly) Bug in NEON ArgMinMax
Change-Id: Id5214de8d679c514132efe81489c637613573224
Reviewed-on: https://review.mlplatform.org/516
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/core/NEON/kernels/NEReductionOperationKernel.cpp')
| -rw-r--r-- | src/core/NEON/kernels/NEReductionOperationKernel.cpp | 11 |
1 files changed, 7 insertions, 4 deletions
diff --git a/src/core/NEON/kernels/NEReductionOperationKernel.cpp b/src/core/NEON/kernels/NEReductionOperationKernel.cpp index 64e3cfe404..d88717a396 100644 --- a/src/core/NEON/kernels/NEReductionOperationKernel.cpp +++ b/src/core/NEON/kernels/NEReductionOperationKernel.cpp @@ -74,10 +74,13 @@ uint32x4x4_t calculate_index(uint32_t idx, uint8x16_t a, uint8x16_t b, uint32x4x { mask_u8 = wrapper::vclt(b, a); } - mask.val[0] = wrapper::vmovl(wrapper::vgetlow(wrapper::vmovl(wrapper::vgetlow(mask_u8)))); - mask.val[1] = wrapper::vmovl(wrapper::vgethigh(wrapper::vmovl(wrapper::vgetlow(mask_u8)))); - mask.val[2] = wrapper::vmovl(wrapper::vgetlow(wrapper::vmovl(wrapper::vgethigh(mask_u8)))); - mask.val[3] = wrapper::vmovl(wrapper::vgethigh(wrapper::vmovl(wrapper::vgethigh(mask_u8)))); + auto wide_u16_1 = wrapper::vorr(vshll_n_u8(wrapper::vgetlow(mask_u8), 8), wrapper::vmovl(wrapper::vgetlow(mask_u8))); + auto wide_u16_2 = wrapper::vorr(vshll_n_u8(wrapper::vgethigh(mask_u8), 8), wrapper::vmovl(wrapper::vgethigh(mask_u8))); + mask.val[0] = wrapper::vorr(vshll_n_u16(wrapper::vgetlow(wide_u16_1), 16), wrapper::vmovl(wrapper::vgetlow(wide_u16_1))); + mask.val[1] = wrapper::vorr(vshll_n_u16(wrapper::vgethigh(wide_u16_1), 16), wrapper::vmovl(wrapper::vgethigh(wide_u16_1))); + mask.val[2] = wrapper::vorr(vshll_n_u16(wrapper::vgetlow(wide_u16_2), 16), wrapper::vmovl(wrapper::vgetlow(wide_u16_2))); + mask.val[3] = wrapper::vorr(vshll_n_u16(wrapper::vgethigh(wide_u16_2), 16), wrapper::vmovl(wrapper::vgethigh(wide_u16_2))); + uint32x4x4_t vec_idx = { { { idx + 0, idx + 1, idx + 2, idx + 3 }, { idx + 4, idx + 5, idx + 6, idx + 7 }, { idx + 8, idx + 9, idx + 10, idx + 11 }, |