File size: 1,972 Bytes
8b7c501 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 |
// Copyright 2022 Google LLC
//
// This source code is licensed under the BSD-style license found in the
// LICENSE file in the root directory of this source tree.
$assert BATCH_TILE % 4 == 0
$assert BATCH_TILE >= 4
$SIMD_TILE = BATCH_TILE // 4
#include <assert.h>
#include <stddef.h>
#include <stdint.h>
#include <arm_neon.h>
#include <xnnpack/vsquareabs.h>
void xnn_cs16_vsquareabs_ukernel__neon_mlal_ld128_x${BATCH_TILE}(
size_t batch,
const int16_t* input,
uint32_t* output) XNN_OOB_READS
{
assert(batch != 0);
assert(batch % (sizeof(int16_t) * 2) == 0);
assert(input != NULL);
assert(output != NULL);
$if BATCH_TILE > 4:
for (; batch >= ${BATCH_TILE * 2} * sizeof(int16_t); batch -= ${BATCH_TILE * 2} * sizeof(int16_t)) {
$for N in range(SIMD_TILE):
const int16x4x2_t vi${N} = vld2_s16(input); input += 8;
$for N in range(SIMD_TILE):
int32x4_t vacc${N} = vmull_s16(vi${N}.val[0], vi${N}.val[0]);
$for N in range(SIMD_TILE):
vacc${N} = vmlal_s16(vacc${N}, vi${N}.val[1], vi${N}.val[1]);
$for N in range(SIMD_TILE):
vst1q_u32(output, vreinterpretq_u32_s32(vacc${N})); output += 4;
}
for (; batch >= 8 * sizeof(int16_t); batch -= 8 * sizeof(int16_t)) {
const int16x4x2_t vi = vld2_s16(input); input += 8;
int32x4_t vacc = vmull_s16(vi.val[0], vi.val[0]);
vacc = vmlal_s16(vacc, vi.val[1], vi.val[1]);
vst1q_u32(output, vreinterpretq_u32_s32(vacc)); output += 4;
}
if XNN_LIKELY(batch != 0) {
const int16x4x2_t vi = vld2_s16(input);
int32x4_t vacc = vmull_s16(vi.val[0], vi.val[0]);
vacc = vmlal_s16(vacc, vi.val[1], vi.val[1]);
uint32x2_t vacc_lo = vreinterpret_u32_s32(vget_low_s32(vacc));
if (batch & (4 * sizeof(int16_t))) {
vst1_u32(output, vacc_lo); output += 2;
vacc_lo = vreinterpret_u32_s32(vget_high_s32(vacc));
}
if (batch & (2 * sizeof(int16_t))) {
vst1_lane_u32(output, vacc_lo, 0);
}
}
}
|