void test_vld1Q_lanes32 (void)
{
int32x4_t out_int32x4_t;
int32x4_t arg1_int32x4_t;
out_int32x4_t = vld1q_lane_s32 (0, arg1_int32x4_t, 1);
}
void test_ld1st1(int8x8_t small, int8x16_t big, void *addr) {
vld1_lane_s8(addr, small, 7);
vld1_lane_s16(addr, small, 3);
vld1_lane_s32(addr, small, 1);
vld1_lane_s64(addr, small, 0);
vld1q_lane_s8(addr, big, 15);
vld1q_lane_s16(addr, big, 7);
vld1q_lane_s32(addr, big, 3);
vld1q_lane_s64(addr, big, 1);
vld1_lane_s8(addr, small, 8); // expected-error {{argument should be a value from 0 to 7}}
vld1_lane_s16(addr, small, 4); // expected-error {{argument should be a value from 0 to 3}}
vld1_lane_s32(addr, small, 2); // expected-error {{argument should be a value from 0 to 1}}
vld1_lane_s64(addr, small, 1); // expected-error {{argument should be a value from 0 to 0}}
vld1q_lane_s8(addr, big, 16); // expected-error {{argument should be a value from 0 to 15}}
vld1q_lane_s16(addr, big, 8); // expected-error {{argument should be a value from 0 to 7}}
vld1q_lane_s32(addr, big, 4); // expected-error {{argument should be a value from 0 to 3}}
vld1q_lane_s64(addr, big, 2); // expected-error {{argument should be a value from 0 to 1}}
vst1_lane_s8(addr, small, 7);
vst1_lane_s16(addr, small, 3);
vst1_lane_s32(addr, small, 1);
vst1_lane_s64(addr, small, 0);
vst1q_lane_s8(addr, big, 15);
vst1q_lane_s16(addr, big, 7);
vst1q_lane_s32(addr, big, 3);
vst1q_lane_s64(addr, big, 1);
vst1_lane_s8(addr, small, 8); // expected-error {{argument should be a value from 0 to 7}}
vst1_lane_s16(addr, small, 4); // expected-error {{argument should be a value from 0 to 3}}
vst1_lane_s32(addr, small, 2); // expected-error {{argument should be a value from 0 to 1}}
vst1_lane_s64(addr, small, 1); // expected-error {{argument should be a value from 0 to 0}}
vst1q_lane_s8(addr, big, 16); // expected-error {{argument should be a value from 0 to 15}}
vst1q_lane_s16(addr, big, 8); // expected-error {{argument should be a value from 0 to 7}}
vst1q_lane_s32(addr, big, 4); // expected-error {{argument should be a value from 0 to 3}}
vst1q_lane_s64(addr, big, 2); // expected-error {{argument should be a value from 0 to 1}}
}
void WebRtcIsacfix_AllpassFilter2FixDec16Neon(
int16_t* data_ch1, // Input and output in channel 1, in Q0
int16_t* data_ch2, // Input and output in channel 2, in Q0
const int16_t* factor_ch1, // Scaling factor for channel 1, in Q15
const int16_t* factor_ch2, // Scaling factor for channel 2, in Q15
const int length, // Length of the data buffers
int32_t* filter_state_ch1, // Filter state for channel 1, in Q16
int32_t* filter_state_ch2) { // Filter state for channel 2, in Q16
assert(length % 2 == 0);
int n = 0;
int16x4_t factorv;
int16x4_t datav;
int32x4_t statev;
int32x2_t tmp;
// Load factor_ch1 and factor_ch2.
tmp = vld1_dup_s32((int32_t*)factor_ch1);
tmp = vld1_lane_s32((int32_t*)factor_ch2, tmp, 1);
factorv = vreinterpret_s16_s32(tmp);
// Load filter_state_ch1[0] and filter_state_ch2[0].
statev = vld1q_dup_s32(filter_state_ch1);
statev = vld1q_lane_s32(filter_state_ch2, statev, 2);
// Loop unrolling preprocessing.
int32x4_t a;
int16x4_t tmp1, tmp2;
// Load data_ch1[0] and data_ch2[0].
datav = vld1_dup_s16(data_ch1);
datav = vld1_lane_s16(data_ch2, datav, 2);
a = vqdmlal_s16(statev, datav, factorv);
tmp1 = vshrn_n_s32(a, 16);
// Update filter_state_ch1[0] and filter_state_ch2[0].
statev = vqdmlsl_s16(vshll_n_s16(datav, 16), tmp1, factorv);
// Load filter_state_ch1[1] and filter_state_ch2[1].
statev = vld1q_lane_s32(filter_state_ch1 + 1, statev, 1);
statev = vld1q_lane_s32(filter_state_ch2 + 1, statev, 3);
// Load data_ch1[1] and data_ch2[1].
tmp1 = vld1_lane_s16(data_ch1 + 1, tmp1, 1);
tmp1 = vld1_lane_s16(data_ch2 + 1, tmp1, 3);
datav = vrev32_s16(tmp1);
// Loop unrolling processing.
for (n = 0; n < length - 2; n += 2) {
a = vqdmlal_s16(statev, datav, factorv);
tmp1 = vshrn_n_s32(a, 16);
// Store data_ch1[n] and data_ch2[n].
vst1_lane_s16(data_ch1 + n, tmp1, 1);
vst1_lane_s16(data_ch2 + n, tmp1, 3);
// Update filter_state_ch1[0], filter_state_ch1[1]
// and filter_state_ch2[0], filter_state_ch2[1].
statev = vqdmlsl_s16(vshll_n_s16(datav, 16), tmp1, factorv);
// Load data_ch1[n + 2] and data_ch2[n + 2].
tmp1 = vld1_lane_s16(data_ch1 + n + 2, tmp1, 1);
tmp1 = vld1_lane_s16(data_ch2 + n + 2, tmp1, 3);
datav = vrev32_s16(tmp1);
a = vqdmlal_s16(statev, datav, factorv);
tmp2 = vshrn_n_s32(a, 16);
// Store data_ch1[n + 1] and data_ch2[n + 1].
vst1_lane_s16(data_ch1 + n + 1, tmp2, 1);
vst1_lane_s16(data_ch2 + n + 1, tmp2, 3);
// Update filter_state_ch1[0], filter_state_ch1[1]
// and filter_state_ch2[0], filter_state_ch2[1].
statev = vqdmlsl_s16(vshll_n_s16(datav, 16), tmp2, factorv);
// Load data_ch1[n + 3] and data_ch2[n + 3].
tmp2 = vld1_lane_s16(data_ch1 + n + 3, tmp2, 1);
tmp2 = vld1_lane_s16(data_ch2 + n + 3, tmp2, 3);
datav = vrev32_s16(tmp2);
}
// Loop unrolling post-processing.
a = vqdmlal_s16(statev, datav, factorv);
tmp1 = vshrn_n_s32(a, 16);
// Store data_ch1[n] and data_ch2[n].
vst1_lane_s16(data_ch1 + n, tmp1, 1);
vst1_lane_s16(data_ch2 + n, tmp1, 3);
// Update filter_state_ch1[0], filter_state_ch1[1]
// and filter_state_ch2[0], filter_state_ch2[1].
statev = vqdmlsl_s16(vshll_n_s16(datav, 16), tmp1, factorv);
// Store filter_state_ch1[0] and filter_state_ch2[0].
vst1q_lane_s32(filter_state_ch1, statev, 0);
vst1q_lane_s32(filter_state_ch2, statev, 2);
datav = vrev32_s16(tmp1);
a = vqdmlal_s16(statev, datav, factorv);
tmp2 = vshrn_n_s32(a, 16);
// Store data_ch1[n + 1] and data_ch2[n + 1].
vst1_lane_s16(data_ch1 + n + 1, tmp2, 1);
vst1_lane_s16(data_ch2 + n + 1, tmp2, 3);
// Update filter_state_ch1[1] and filter_state_ch2[1].
statev = vqdmlsl_s16(vshll_n_s16(datav, 16), tmp2, factorv);
// Store filter_state_ch1[1] and filter_state_ch2[1].
vst1q_lane_s32(filter_state_ch1 + 1, statev, 1);
vst1q_lane_s32(filter_state_ch2 + 1, statev, 3);
}
