search for: vadd_f32

...ues per cycle */ s/cycle/iteration/ (here and below). In performance-critical code when you say cycle I think machine cycle, and NEON definitely can't process 16 floats in one of those. > + while (len >= 16) { > + /* Accumulate results into single float */ > + tv.val[0] = vadd_f32(vget_low_f32(SUMM), vget_high_f32(SUMM)); > + tv = vtrn_f32(tv.val[0], ZERO); > + tv.val[0] = vadd_f32(tv.val[0], tv.val[1]); > + > + vst1_lane_f32(&sumi, tv.val[0], 0); Accessing tv.val[0] and tv.val[1] directly seems to send these values through the stack, e.g., f4:...

...= vld1q_f32(ai); + ai += 4; + Fout_4[1] = vld1q_f32(ai); + ai += 4; + Fout_2[0] = vget_low_f32(Fout_4[0]); + Fout_2[1] = vget_high_f32(Fout_4[0]); + Fout_2[2] = vget_low_f32(Fout_4[1]); + Fout_2[3] = vget_high_f32(Fout_4[1]); + + scratch_2[0] = vsub_f32(Fout_2[0], Fout_2[2]); + Fout_2[0] = vadd_f32(Fout_2[0], Fout_2[2]); + scratch_2[1] = vadd_f32(Fout_2[1], Fout_2[3]); + Fout_2[2] = vsub_f32(Fout_2[0], scratch_2[1]); + Fout_2[0] = vadd_f32(Fout_2[0], scratch_2[1]); + scratch_2[1] = vsub_f32(Fout_2[1], Fout_2[3]); + + scratch_2[1] = vrev64_f32(scratch_2[1]); + /* scratch_2[1] *= (1, -1)...

Hello, This patch introduces ARM NEON Intrinsics to optimize kf_bfly4 routine in celt part of libopus. Using NEON optimized kf_bfly4(_neon) routine helped improve performance of opus_fft_impl function by about 21.4%. The end use case was decoding a music opus ogg file. The end use case saw performance improvement of about 4.47%. This patch has 2 components i. Actual neon code to improve

Hi, Optimizes celt_pitch_xcorr for floating point. Changes from RFCv1: - Rebased on top of commit aad281878: Fix celt_pitch_xcorr_c signature. which got rid of ugly code around CELT_PITCH_XCORR_IMPL passing of "arch" parameter. - Unified with --enable-intrinsics used by x86 - Modified algorithm to be more in-line with algorithm in celt_pitch_xcorr_arm.s Viswanath Puttagunta

...-= 8; + } + + /* Work on 4 values per cycle */ + while (len >= 4) { + XX[0] = vld1q_f32(xi); + xi += 4; + YY[0] = vld1q_f32(yi); + yi += 4; + SUMM = vmlaq_f32(SUMM, YY[0], XX[0]); + len -= 4; + } + /* Accumulate results into single float */ + tv.val[0] = vadd_f32(vget_low_f32(SUMM), vget_high_f32(SUMM)); + tv = vtrn_f32(tv.val[0], ZERO); + tv.val[0] = vadd_f32(tv.val[0], tv.val[1]); + + vst1_lane_f32(&sumi, tv.val[0], 0); + for (i = 0; i < len; i++) + sumi += xi[i] * yi[i]; + *sum = sumi; +} + +void celt_pitch_xcorr_float_neon(const op...

Should call celt_inner_prod(). --- celt/bands.c | 7 ++++--- celt/bands.h | 2 +- celt/celt_encoder.c | 6 +++--- celt/pitch.c | 2 +- src/opus_multistream_encoder.c | 2 +- 5 files changed, 10 insertions(+), 9 deletions(-) diff --git a/celt/bands.c b/celt/bands.c index bbe8a4c..1ab24aa 100644 --- a/celt/bands.c +++ b/celt/bands.c

From: Viswanath Puttagunta <viswanath.puttagunta at linaro.org> Hi, Optimizes celt_pitch_xcorr for floating point. Changes from RFCv1: - Rebased on top of commit aad281878: Fix celt_pitch_xcorr_c signature. which got rid of ugly code around CELT_PITCH_XCORR_IMPL passing of "arch" parameter. - Unified with --enable-intrinsics used by x86 - Modified algorithm to be more

Hi, Optimizes celt_pitch_xcorr for ARM NEON floating point. Changes from RFCv3: - celt_neon_intr.c - removed warnings due to not having constant pointers - Put simpler loop to take care of corner cases. Unrolling using intrinsics was not really mapping well to what was done in celt_pitch_xcorr_arm.s - Makefile.am Removed explicit -O3 optimization - test_unit_mathops.c,

...+ if (len >= 2) { + /* While at it, consume 2 more values if available */ + XX_2 = vld1_f32(xi); + xi += 2; + YY_2 = vld1_f32(yi); + yi += 2; + SUMM_2[0] = vmla_f32(SUMM_2[0], YY_2, XX_2); + len -= 2; + } + SUMM_2[1] = vget_high_f32(SUMM); + SUMM_2[0] = vadd_f32(SUMM_2[0], SUMM_2[1]); + SUMM_2[0] = vpadd_f32(SUMM_2[0], SUMM_2[0]); + /* Ok, now we have result accumulated in SUMM_2[0].0 */ + + if (len > 0) { + /* Case when you have one value left */ + XX_2 = vld1_dup_f32(xi); + YY_2 = vld1_dup_f32(yi); + SUMM_2[0] = vmla_f32(SUMM...

...+ if (len >= 2) { + /* While at it, consume 2 more values if available */ + XX_2 = vld1_f32(xi); + xi += 2; + YY_2 = vld1_f32(yi); + yi += 2; + SUMM_2[0] = vmla_f32(SUMM_2[0], YY_2, XX_2); + len -= 2; + } + SUMM_2[1] = vget_high_f32(SUMM); + SUMM_2[0] = vadd_f32(SUMM_2[0], SUMM_2[1]); + SUMM_2[0] = vpadd_f32(SUMM_2[0], SUMM_2[0]); + /* Ok, now we have result accumulated in SUMM_2[0].0 */ + + if (len > 0) { + /* Case when you have one value left */ + XX_2 = vld1_dup_f32(xi); + YY_2 = vld1_dup_f32(yi); + SUMM_2[0] = vmla_f32(SUMM...

Hi, Optimizes celt_pitch_xcorr for floating point. Changes from RFCv2: - Changes recommended by Timothy for celt_neon_intr.c everything except, left the unrolled loop still unrolled - configure.ac - use AC_LINK_IFELSE instead of AC_COMPILE_IFELSE - Moved compile flags into Makefile.am - OPUS_ARM_NEON_INR --> typo --> OPUS_ARM_NEON_INTR Viswanath Puttagunta (1): armv7:

...consume 2 more values if available */ > + XX_2 = vld1_f32(xi); > + xi += 2; > + YY_2 = vld1_f32(yi); > + yi += 2; > + SUMM_2[0] = vmla_f32(SUMM_2[0], YY_2, XX_2); > + len -= 2; > + } > + SUMM_2[1] = vget_high_f32(SUMM); > + SUMM_2[0] = vadd_f32(SUMM_2[0], SUMM_2[1]); > + SUMM_2[0] = vpadd_f32(SUMM_2[0], SUMM_2[0]); > + /* Ok, now we have result accumulated in SUMM_2[0].0 */ > + > + if (len > 0) { > + /* Case when you have one value left */ > + XX_2 = vld1_dup_f32(xi); > + YY_2 = vld1_dup_f32(yi...