Skip to content

Commit

Permalink
who knows
Browse files Browse the repository at this point in the history
  • Loading branch information
@argilo
argilo committed Feb 1, 2024
1 parent 4fbfaa6 commit e5efa34
Showing 1 changed file with 17 additions and 17 deletions.
34 changes: 17 additions & 17 deletions kernels/volk/volk_8u_x4_conv_k7_r2_8u.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -669,18 +669,17 @@ static inline void volk_8u_x4_conv_k7_r2_8u_neonspiral(unsigned char* Y,
high_bits = vreinterpretq_u16_u8(vshrq_n_u8(d9, 7));
paired16 = vreinterpretq_u32_u16(vsraq_n_u16(high_bits, high_bits, 6));
paired32_new = vreinterpretq_u8_u32(vsraq_n_u32(paired16, paired16, 12));
s20 = ((unsigned int)vgetq_lane_u8(paired32_new, 0) << 9)
| ((unsigned int)vgetq_lane_u8(paired32_new, 4) << 1)
| ((unsigned int)vgetq_lane_u8(paired32_new, 8) << 25)
| ((unsigned int)vgetq_lane_u8(paired32_new, 12) << 17);
printf("%08x\n", s20);
s20 = ((unsigned int)vgetq_lane_u8(paired32_new, 0) << 1)
| ((unsigned int)vgetq_lane_u8(paired32_new, 4) << 9)
| ((unsigned int)vgetq_lane_u8(paired32_new, 8) << 17)
| ((unsigned int)vgetq_lane_u8(paired32_new, 12) << 25);
high_bits = vreinterpretq_u16_u8(vshrq_n_u8(d10, 7));
paired16 = vreinterpretq_u32_u16(vsraq_n_u16(high_bits, high_bits, 6));
paired32_new = vreinterpretq_u8_u32(vsraq_n_u32(paired16, paired16, 12));
s20 |= ((unsigned int)vgetq_lane_u8(paired32_new, 0) << 8)
| ((unsigned int)vgetq_lane_u8(paired32_new, 4) << 0)
| ((unsigned int)vgetq_lane_u8(paired32_new, 8) << 24)
| ((unsigned int)vgetq_lane_u8(paired32_new, 12) << 16);
s20 |= ((unsigned int)vgetq_lane_u8(paired32_new, 0) << 0)
| ((unsigned int)vgetq_lane_u8(paired32_new, 4) << 8)
| ((unsigned int)vgetq_lane_u8(paired32_new, 8) << 16)
| ((unsigned int)vgetq_lane_u8(paired32_new, 12) << 24);
printf("%08x\n", s20);
a91 = ((unsigned int*)dec);
a92 = (4 * i9);
Expand Down Expand Up @@ -717,17 +716,18 @@ printf("%08x\n", s20);
high_bits = vreinterpretq_u16_u8(vshrq_n_u8(d11, 7));
paired16 = vreinterpretq_u32_u16(vsraq_n_u16(high_bits, high_bits, 6));
paired32_new = vreinterpretq_u8_u32(vsraq_n_u32(paired16, paired16, 12));
s26 = ((unsigned int)vgetq_lane_u8(paired32_new, 0) << 9)
| ((unsigned int)vgetq_lane_u8(paired32_new, 4) << 1)
| ((unsigned int)vgetq_lane_u8(paired32_new, 8) << 25)
| ((unsigned int)vgetq_lane_u8(paired32_new, 12) << 17);
s26 = ((unsigned int)vgetq_lane_u8(paired32_new, 0) << 1)
| ((unsigned int)vgetq_lane_u8(paired32_new, 4) << 9)
| ((unsigned int)vgetq_lane_u8(paired32_new, 8) << 17)
| ((unsigned int)vgetq_lane_u8(paired32_new, 12) << 25);
high_bits = vreinterpretq_u16_u8(vshrq_n_u8(d12, 7));
paired16 = vreinterpretq_u32_u16(vsraq_n_u16(high_bits, high_bits, 6));
paired32_new = vreinterpretq_u8_u32(vsraq_n_u32(paired16, paired16, 12));
s26 |= ((unsigned int)vgetq_lane_u8(paired32_new, 0) << 8)
| ((unsigned int)vgetq_lane_u8(paired32_new, 4) << 0)
| ((unsigned int)vgetq_lane_u8(paired32_new, 8) << 24)
| ((unsigned int)vgetq_lane_u8(paired32_new, 12) << 16);
s26 |= ((unsigned int)vgetq_lane_u8(paired32_new, 0) << 0)
| ((unsigned int)vgetq_lane_u8(paired32_new, 4) << 8)
| ((unsigned int)vgetq_lane_u8(paired32_new, 8) << 16)
| ((unsigned int)vgetq_lane_u8(paired32_new, 12) << 24);
printf("%08x\n", s26);
a110 = (a93 + 1);
*(a110) = s26;
s28 = vzip1q_u8(a108, a109);
Expand Down

0 comments on commit e5efa34

Please sign in to comment.