diff options
Diffstat (limited to 'src/core/NEON/kernels/arm_gemm/transforms/a64_block16_interleave4_8bit.hpp')
-rw-r--r-- | src/core/NEON/kernels/arm_gemm/transforms/a64_block16_interleave4_8bit.hpp | 12 |
1 files changed, 5 insertions, 7 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/transforms/a64_block16_interleave4_8bit.hpp b/src/core/NEON/kernels/arm_gemm/transforms/a64_block16_interleave4_8bit.hpp index 6b742c8776..9b6f4de543 100644 --- a/src/core/NEON/kernels/arm_gemm/transforms/a64_block16_interleave4_8bit.hpp +++ b/src/core/NEON/kernels/arm_gemm/transforms/a64_block16_interleave4_8bit.hpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2017-2018 ARM Limited. + * Copyright (c) 2017-2020 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -38,8 +38,8 @@ void TransformImpl<4, 16, false, 1, 1, false>::Transform(T *out, const T *in, in uint8_t zerobuff[16] = { 0 }; - for (uint64_t y = y0 ; y < static_cast<uint64_t>(ymax) ; y+=4) { - const uint8_t *inptr0 = inptr + y * ldin + k0; + for (int y=y0; y<ymax; y+=4) { + const uint8_t *inptr0 = inptr + static_cast<intptr_t>(y) * ldin + k0; const uint8_t *inptr1 = inptr0 + ldin; const uint8_t *inptr2 = inptr1 + ldin; const uint8_t *inptr3 = inptr2 + ldin; @@ -52,9 +52,8 @@ void TransformImpl<4, 16, false, 1, 1, false>::Transform(T *out, const T *in, in int x=(kmax-k0); for (;x>15;x-=16) { /* Cope with ragged cases by copying from a buffer of zeroes instead */ - if ((y + 3) >= static_cast<uint64_t>(ymax)) { + if ((y + 3) >= ymax) { switch ((y + 3) - ymax) { - /* Everything falls through in here */ case 2: inptr1 = zerobuff; // fall through @@ -90,9 +89,8 @@ void TransformImpl<4, 16, false, 1, 1, false>::Transform(T *out, const T *in, in if (x>0) { /* Need to duplicate this here, in case we didn't run the main loop. */ - if ((y + 3) >= static_cast<uint64_t>(ymax)) { + if ((y + 3) >= ymax) { switch ((y + 3) - ymax) { - /* Everything falls through in here */ case 2: inptr1 = zerobuff; // fall through |