void jsimd_quantize_altivec()

in simd/powerpc/jquanti-altivec.c [116:250]

• 122 lines of code
• 1 McCabe index (conditional complexity)

void jsimd_quantize_altivec(JCOEFPTR coef_block, DCTELEM *divisors,
                            DCTELEM *workspace)
{
  __vector short row0, row1, row2, row3, row4, row5, row6, row7,
    row0s, row1s, row2s, row3s, row4s, row5s, row6s, row7s,
    corr0, corr1, corr2, corr3, corr4, corr5, corr6, corr7,
    recip0, recip1, recip2, recip3, recip4, recip5, recip6, recip7,
    scale0, scale1, scale2, scale3, scale4, scale5, scale6, scale7;
  __vector unsigned int tmpe, tmpo;

  /* Constants */
  __vector unsigned short pw_word_bit_m1 = { __8X(WORD_BIT - 1) };
#ifdef __BIG_ENDIAN__
  __vector unsigned char shift_pack_index =
    {  0,  1, 16, 17,  4,  5, 20, 21,  8,  9, 24, 25, 12, 13, 28, 29 };
#else
  __vector unsigned char shift_pack_index =
    {  2,  3, 18, 19,  6,  7, 22, 23, 10, 11, 26, 27, 14, 15, 30, 31 };
#endif

  row0 = vec_ld(0, workspace);
  row1 = vec_ld(16, workspace);
  row2 = vec_ld(32, workspace);
  row3 = vec_ld(48, workspace);
  row4 = vec_ld(64, workspace);
  row5 = vec_ld(80, workspace);
  row6 = vec_ld(96, workspace);
  row7 = vec_ld(112, workspace);

  /* Branch-less absolute value */
  row0s = vec_sra(row0, pw_word_bit_m1);
  row1s = vec_sra(row1, pw_word_bit_m1);
  row2s = vec_sra(row2, pw_word_bit_m1);
  row3s = vec_sra(row3, pw_word_bit_m1);
  row4s = vec_sra(row4, pw_word_bit_m1);
  row5s = vec_sra(row5, pw_word_bit_m1);
  row6s = vec_sra(row6, pw_word_bit_m1);
  row7s = vec_sra(row7, pw_word_bit_m1);
  row0 = vec_xor(row0, row0s);
  row1 = vec_xor(row1, row1s);
  row2 = vec_xor(row2, row2s);
  row3 = vec_xor(row3, row3s);
  row4 = vec_xor(row4, row4s);
  row5 = vec_xor(row5, row5s);
  row6 = vec_xor(row6, row6s);
  row7 = vec_xor(row7, row7s);
  row0 = vec_sub(row0, row0s);
  row1 = vec_sub(row1, row1s);
  row2 = vec_sub(row2, row2s);
  row3 = vec_sub(row3, row3s);
  row4 = vec_sub(row4, row4s);
  row5 = vec_sub(row5, row5s);
  row6 = vec_sub(row6, row6s);
  row7 = vec_sub(row7, row7s);

  corr0 = vec_ld(DCTSIZE2 * 2, divisors);
  corr1 = vec_ld(DCTSIZE2 * 2 + 16, divisors);
  corr2 = vec_ld(DCTSIZE2 * 2 + 32, divisors);
  corr3 = vec_ld(DCTSIZE2 * 2 + 48, divisors);
  corr4 = vec_ld(DCTSIZE2 * 2 + 64, divisors);
  corr5 = vec_ld(DCTSIZE2 * 2 + 80, divisors);
  corr6 = vec_ld(DCTSIZE2 * 2 + 96, divisors);
  corr7 = vec_ld(DCTSIZE2 * 2 + 112, divisors);

  row0 = vec_add(row0, corr0);
  row1 = vec_add(row1, corr1);
  row2 = vec_add(row2, corr2);
  row3 = vec_add(row3, corr3);
  row4 = vec_add(row4, corr4);
  row5 = vec_add(row5, corr5);
  row6 = vec_add(row6, corr6);
  row7 = vec_add(row7, corr7);

  recip0 = vec_ld(0, divisors);
  recip1 = vec_ld(16, divisors);
  recip2 = vec_ld(32, divisors);
  recip3 = vec_ld(48, divisors);
  recip4 = vec_ld(64, divisors);
  recip5 = vec_ld(80, divisors);
  recip6 = vec_ld(96, divisors);
  recip7 = vec_ld(112, divisors);

  MULTIPLY(row0, recip0, row0);
  MULTIPLY(row1, recip1, row1);
  MULTIPLY(row2, recip2, row2);
  MULTIPLY(row3, recip3, row3);
  MULTIPLY(row4, recip4, row4);
  MULTIPLY(row5, recip5, row5);
  MULTIPLY(row6, recip6, row6);
  MULTIPLY(row7, recip7, row7);

  scale0 = vec_ld(DCTSIZE2 * 4, divisors);
  scale1 = vec_ld(DCTSIZE2 * 4 + 16, divisors);
  scale2 = vec_ld(DCTSIZE2 * 4 + 32, divisors);
  scale3 = vec_ld(DCTSIZE2 * 4 + 48, divisors);
  scale4 = vec_ld(DCTSIZE2 * 4 + 64, divisors);
  scale5 = vec_ld(DCTSIZE2 * 4 + 80, divisors);
  scale6 = vec_ld(DCTSIZE2 * 4 + 96, divisors);
  scale7 = vec_ld(DCTSIZE2 * 4 + 112, divisors);

  MULTIPLY(row0, scale0, row0);
  MULTIPLY(row1, scale1, row1);
  MULTIPLY(row2, scale2, row2);
  MULTIPLY(row3, scale3, row3);
  MULTIPLY(row4, scale4, row4);
  MULTIPLY(row5, scale5, row5);
  MULTIPLY(row6, scale6, row6);
  MULTIPLY(row7, scale7, row7);

  row0 = vec_xor(row0, row0s);
  row1 = vec_xor(row1, row1s);
  row2 = vec_xor(row2, row2s);
  row3 = vec_xor(row3, row3s);
  row4 = vec_xor(row4, row4s);
  row5 = vec_xor(row5, row5s);
  row6 = vec_xor(row6, row6s);
  row7 = vec_xor(row7, row7s);
  row0 = vec_sub(row0, row0s);
  row1 = vec_sub(row1, row1s);
  row2 = vec_sub(row2, row2s);
  row3 = vec_sub(row3, row3s);
  row4 = vec_sub(row4, row4s);
  row5 = vec_sub(row5, row5s);
  row6 = vec_sub(row6, row6s);
  row7 = vec_sub(row7, row7s);

  vec_st(row0, 0, coef_block);
  vec_st(row1, 16, coef_block);
  vec_st(row2, 32, coef_block);
  vec_st(row3, 48, coef_block);
  vec_st(row4, 64, coef_block);
  vec_st(row5, 80, coef_block);
  vec_st(row6, 96, coef_block);
  vec_st(row7, 112, coef_block);
}